@bitkyc08/opencodex 2.91.0 → 2.93.0-preview.20261027
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bin/ocx.mjs +47 -8
- package/gui/dist/assets/App-CJZoq-uF.js +58 -0
- package/gui/dist/assets/{Tray-BuLy7hL1.js → Tray-BWQPcgZn.js} +1 -1
- package/gui/dist/assets/index-DRe8Hhfz.css +1 -0
- package/gui/dist/assets/index-Xfct8owH.js +94 -0
- package/gui/dist/assets/quota-summary-DxToX1rV.js +1 -0
- package/gui/dist/index.html +2 -2
- package/gui/dist/provider-icons/workbuddy-color.svg +29 -0
- package/package.json +8 -7
- package/src/adapters/anthropic.ts +28 -5
- package/src/adapters/coding-agent/batch-dispatcher.ts +122 -0
- package/src/adapters/coding-agent/profile.ts +54 -16
- package/src/adapters/coding-agent/turn.ts +30 -2
- package/src/adapters/command-code.ts +114 -7
- package/src/adapters/cursor/native-exec-fs.ts +11 -9
- package/src/adapters/devin/live-models.ts +4 -0
- package/src/adapters/google-vertex-endpoint.ts +9 -1
- package/src/adapters/google-vertex-key-project.ts +223 -0
- package/src/adapters/google-wire-compiler.ts +1 -2
- package/src/adapters/google.ts +13 -3
- package/src/adapters/kiro/reasoning.ts +2 -0
- package/src/adapters/kiro-refusal.ts +6 -1
- package/src/adapters/kiro-wire.ts +1 -1
- package/src/adapters/openai-chat/messages.ts +8 -2
- package/src/adapters/openai-chat/passthrough.ts +35 -14
- package/src/adapters/openai-chat/reasoning-replay.ts +27 -16
- package/src/adapters/openai-chat/reasoning-wire.ts +117 -8
- package/src/adapters/openai-chat/required-tool-result.ts +64 -0
- package/src/adapters/openai-chat/tool-name-registry.ts +3 -1
- package/src/adapters/openai-chat.ts +9 -10
- package/src/adapters/openai-responses/passthrough.ts +7 -4
- package/src/adapters/openai-responses/reasoning.ts +13 -0
- package/src/adapters/openai-responses/request-strips.ts +60 -0
- package/src/adapters/qoder/adapter.ts +129 -8
- package/src/adapters/qoder/scaffold-guard.ts +5 -0
- package/src/adapters/reasoning-controls.ts +16 -0
- package/src/adapters/registry.ts +5 -0
- package/src/adapters/tool-call-id.ts +1 -3
- package/src/adapters/workbuddy-schema.ts +61 -0
- package/src/adapters/workbuddy.ts +68 -0
- package/src/bridge/errors.ts +27 -0
- package/src/bridge/response-json.ts +1 -1
- package/src/bridge/sse.ts +1 -1
- package/src/bridge.ts +1 -1
- package/src/chat/inbound.ts +4 -1
- package/src/chat/outbound.ts +20 -6
- package/src/chat/reasoning-details.ts +28 -5
- package/src/claude/desktop-first-party.ts +54 -4
- package/src/claude/desktop-picker.ts +11 -8
- package/src/claude/first-party-settings.ts +3 -1
- package/src/claude/inbound.ts +2 -1
- package/src/claude/intercept/local-ca.ts +2 -1
- package/src/claude/intercept/picker-ca-cleanup.ts +4 -2
- package/src/claude/intercept/picker-ca-startup.ts +3 -1
- package/src/claude/intercept/picker-models.ts +6 -3
- package/src/claude/intercept/picker-runtime.ts +3 -2
- package/src/claude/intercept/runtime.ts +6 -3
- package/src/claude/model-info.ts +31 -8
- package/src/cli/account-api.ts +11 -0
- package/src/cli/capabilities-agents-routing.ts +3 -3
- package/src/cli/capabilities-base.ts +7 -1
- package/src/cli/capabilities-observe-system.ts +10 -0
- package/src/cli/claude-desktop.ts +9 -8
- package/src/cli/claude.ts +3 -2
- package/src/cli/codex-cli-update.ts +8 -0
- package/src/cli/codex-shim-autorestore.ts +9 -0
- package/src/cli/combo-result.ts +13 -0
- package/src/cli/combo-stats.ts +3 -0
- package/src/cli/combo.ts +83 -41
- package/src/cli/cross-home-owner.ts +13 -1
- package/src/cli/dispatch.ts +25 -6
- package/src/cli/ensure-desired-integrations.ts +8 -6
- package/src/cli/export-command.ts +4 -1
- package/src/cli/help.ts +2 -1
- package/src/cli/index.ts +45 -45
- package/src/cli/integration-plan-dto.ts +4 -0
- package/src/cli/launcher-context.ts +49 -2
- package/src/cli/link.ts +45 -32
- package/src/cli/opencode.ts +20 -9
- package/src/cli/provider.ts +2 -1
- package/src/cli/registry.ts +6 -4
- package/src/cli/root.ts +9 -0
- package/src/cli/status.ts +59 -1
- package/src/cli/stop-restore.ts +113 -0
- package/src/cli/sync-expectation.ts +60 -0
- package/src/cli/uninstall-integrations.ts +41 -7
- package/src/cli/usage-compare.ts +68 -12
- package/src/client/connect.ts +40 -3
- package/src/client/link-home.ts +10 -1
- package/src/client/link-ingress.ts +1 -1
- package/src/client/link-readiness.ts +97 -0
- package/src/client/link-relay-transport.ts +79 -11
- package/src/client/link-relay.ts +74 -15
- package/src/client/link-tunnel.ts +45 -18
- package/src/client/machine-api.ts +3 -1
- package/src/client/machine-listener.ts +20 -2
- package/src/client/runtime.ts +20 -2
- package/src/client/sync-expectation.ts +118 -0
- package/src/clients/config-export/agent-clients.ts +96 -0
- package/src/clients/config-export/contracts.ts +25 -7
- package/src/clients/config-export/qoder.ts +38 -23
- package/src/clients/config-export/reasoning-metadata.ts +55 -8
- package/src/clients/config-export.ts +12 -0
- package/src/clients/lazycodex.ts +7 -2
- package/src/clients/reasoning-controls.ts +49 -0
- package/src/codex/account-store.ts +17 -3
- package/src/codex/app-server-restart-service.ts +6 -5
- package/src/codex/auth-api/account-identity.ts +14 -0
- package/src/codex/auth-api/account-list.ts +29 -2
- package/src/codex/auth-api/http.ts +13 -0
- package/src/codex/auth-api/main-account-probe.ts +2 -2
- package/src/codex/auth-api/pool-mode-gate.ts +16 -3
- package/src/codex/auth-api/pool-quota-probe.ts +15 -0
- package/src/codex/auth-context.ts +15 -12
- package/src/codex/catalog/build-entries.ts +22 -0
- package/src/codex/catalog/configured-tiers.ts +70 -0
- package/src/codex/catalog/effort.ts +52 -11
- package/src/codex/catalog/gather-capture.ts +5 -2
- package/src/codex/catalog/heal-observation.ts +155 -13
- package/src/codex/catalog/metadata.ts +6 -0
- package/src/codex/catalog/model-hints.ts +60 -11
- package/src/codex/catalog/model-visibility.ts +3 -2
- package/src/codex/catalog/native-models.ts +18 -3
- package/src/codex/catalog/parsing.ts +10 -8
- package/src/codex/catalog/pinned-models.ts +2 -1
- package/src/codex/catalog/provider-models.ts +18 -13
- package/src/codex/catalog/retained-sync.ts +2 -1
- package/src/codex/catalog/routed-gather.ts +19 -20
- package/src/codex/catalog/write-audit.ts +3 -2
- package/src/codex/catalog-self-heal.ts +43 -23
- package/src/codex/chatgpt-refresh-failure.ts +6 -2
- package/src/codex/cli-install-provenance.ts +418 -0
- package/src/codex/cli-installation-targets.ts +5 -1
- package/src/codex/cli-update-plan.ts +43 -16
- package/src/codex/collision-identity.ts +30 -5
- package/src/codex/convergence.ts +4 -3
- package/src/codex/data/roster-pinned-models.json +177 -1
- package/src/codex/desired-state.ts +6 -1
- package/src/codex/desktop-app/picker-freshness.ts +10 -8
- package/src/codex/desktop-compatibility/installed-build.ts +88 -12
- package/src/codex/desktop-compatibility/runtime.ts +29 -5
- package/src/codex/desktop-compatibility/usage-activation.ts +2 -1
- package/src/codex/desktop-compatibility/usage-controller.ts +9 -1
- package/src/codex/desktop-compatibility/windows-certificate-trust.ts +46 -12
- package/src/codex/inject/config-toml.ts +0 -18
- package/src/codex/inject/plan.ts +18 -3
- package/src/codex/inject/remove.ts +6 -2
- package/src/codex/inject.ts +1 -1
- package/src/codex/journal.ts +61 -1
- package/src/codex/model-entitlements.ts +1 -1
- package/src/codex/native-main-owner.ts +27 -1
- package/src/codex/native-profile-store.ts +2 -1
- package/src/codex/pool-credit-policy.ts +35 -1
- package/src/codex/pool-refresh-backoff.ts +3 -1
- package/src/codex/quota-types.ts +3 -0
- package/src/codex/quota.ts +24 -4
- package/src/codex/role-auto-assign.ts +6 -2
- package/src/codex/subagent-defaults.ts +49 -4
- package/src/codex/user-identity.ts +35 -6
- package/src/codex/windows-current-sid.ts +162 -0
- package/src/combos/failover.ts +17 -2
- package/src/combos/jev-decision-contract.ts +22 -1
- package/src/combos/jev-quota-activation.ts +5 -1
- package/src/combos/jev-quota-route.ts +1 -1
- package/src/combos/jev-quota.ts +9 -1
- package/src/combos/jev-service-exchange.ts +3 -1
- package/src/combos/jev-session-pin-config.ts +51 -0
- package/src/combos/jev.ts +101 -9
- package/src/combos/resolve.ts +45 -1
- package/src/combos/types.ts +43 -0
- package/src/config/load-degrade.ts +5 -0
- package/src/config/paths.ts +19 -3
- package/src/config/pending-teardown.ts +22 -1
- package/src/config/provider-validation.ts +22 -4
- package/src/config/removed-opencode-free.ts +11 -0
- package/src/config/schema/leaf-validators.ts +6 -0
- package/src/generated/compatibility-version.json +467 -311
- package/src/generated/model-metadata.ts +5 -5
- package/src/images/loop.ts +16 -2
- package/src/integrations/catalog-refresh.ts +1 -1
- package/src/integrations/hermes-profiles.ts +251 -0
- package/src/integrations/merge.ts +8 -0
- package/src/integrations/mutation-plan.ts +19 -4
- package/src/integrations/registry.ts +5 -0
- package/src/integrations/state.ts +36 -1
- package/src/integrations/writer.ts +18 -4
- package/src/lib/bun-path-runtime.mjs +2 -1
- package/src/lib/decision-quota-sink.ts +10 -0
- package/src/lib/errors.ts +62 -0
- package/src/lib/eventstream-decoder.ts +81 -37
- package/src/lib/opaque-proxy-tunnel.ts +32 -8
- package/src/lib/pinned-http.ts +35 -2
- package/src/lib/process-control.ts +13 -1
- package/src/lib/socks5-fetch.ts +59 -64
- package/src/lib/socks5-upload-waiter.ts +50 -0
- package/src/lib/sse-decoder.ts +7 -2
- package/src/lib/state-store-sweeper.ts +40 -14
- package/src/lib/translator-budget.ts +4 -4
- package/src/lib/upstream-retry.ts +4 -1
- package/src/lib/windows-native-principal-worker.ts +14 -0
- package/src/lib/windows-native-principal.ts +188 -0
- package/src/lib/windows-secret-acl.ts +39 -3
- package/src/lib/windows-user-principal.ts +62 -18
- package/src/link/diagnostics.ts +193 -0
- package/src/link/relay-auth.ts +6 -0
- package/src/link/ssh-config.ts +35 -3
- package/src/link/status-projection.ts +39 -0
- package/src/link/supervisor.ts +45 -12
- package/src/link/tunnel-history.ts +109 -0
- package/src/link/tunnel-state.ts +80 -17
- package/src/messaging/remote-files.ts +24 -1
- package/src/oauth/generic-account-failover.ts +91 -7
- package/src/oauth/index.ts +15 -0
- package/src/oauth/model-discovery-auth.ts +5 -0
- package/src/oauth/store.ts +4 -1
- package/src/oauth/token-guardian.ts +7 -0
- package/src/oauth/types.ts +9 -0
- package/src/oauth/workbuddy.ts +139 -0
- package/src/providers/account-quota-disk.ts +12 -4
- package/src/providers/antigravity-quota-identity.ts +37 -0
- package/src/providers/base-url-choices.ts +8 -5
- package/src/providers/codebuddy-models.ts +6 -0
- package/src/providers/codex-capacity.ts +3 -1
- package/src/providers/derive.ts +51 -3
- package/src/providers/kiro-models.ts +3 -0
- package/src/providers/mimo-thinking.ts +12 -0
- package/src/providers/model-discovery.ts +2 -0
- package/src/providers/model-rename-fields.ts +2 -0
- package/src/providers/model-rename-migration.ts +10 -0
- package/src/providers/openai-sidecar-credit.ts +2 -2
- package/src/providers/quota/account-cache.ts +3 -2
- package/src/providers/quota/antigravity.ts +25 -14
- package/src/providers/quota/report-cache.ts +13 -4
- package/src/providers/quota/vendor-probes-key.ts +60 -0
- package/src/providers/quota/workbuddy.ts +83 -0
- package/src/providers/quota-decision-antigravity.ts +64 -0
- package/src/providers/quota-decision-publication.ts +11 -0
- package/src/providers/quota-decision-snapshot.ts +21 -4
- package/src/providers/quota-types.ts +2 -0
- package/src/providers/quota.ts +3 -1
- package/src/providers/reasoning-capabilities.ts +76 -0
- package/src/providers/registry/entries-core.ts +23 -12
- package/src/providers/registry/entries-extended.ts +83 -27
- package/src/providers/registry/model-ids.ts +5 -0
- package/src/providers/registry/model-seeds.ts +120 -56
- package/src/providers/registry/types.ts +29 -2
- package/src/providers/resolved-model-policy.ts +30 -10
- package/src/providers/routed-config.ts +24 -1
- package/src/providers/stale-model-roster-migration.ts +123 -18
- package/src/providers/workbuddy-credit-balance.ts +81 -0
- package/src/providers/workbuddy-models.ts +37 -0
- package/src/providers/workbuddy.ts +60 -0
- package/src/remote-control/workspace-pi-runtime.ts +11 -6
- package/src/responses/compaction-summary.ts +45 -0
- package/src/responses/muse-tool-name-alias.ts +1 -2
- package/src/responses/parser.ts +3 -2
- package/src/responses/reasoning-replay-cache.ts +18 -14
- package/src/responses/schema.ts +1 -0
- package/src/responses/thought-signature-replay.ts +10 -2
- package/src/router.ts +2 -0
- package/src/routing/analytics.ts +2 -3
- package/src/routing/capability.ts +8 -2
- package/src/routing/fallback-policy.ts +3 -3
- package/src/routing/history/ingest.ts +28 -3
- package/src/server/auth-cors.ts +4 -2
- package/src/server/chat-completions.ts +21 -1
- package/src/server/chat-native-eligibility.ts +7 -0
- package/src/server/chat-native-reasoning.ts +24 -7
- package/src/server/chat-native.ts +13 -2
- package/src/server/claude-messages.ts +3 -0
- package/src/server/direct-local-http.ts +5 -2
- package/src/server/effort-policy.ts +5 -1
- package/src/server/index/link-relay-sessions.ts +9 -5
- package/src/server/index/live-sideband.ts +18 -15
- package/src/server/index/optional-listeners.ts +4 -0
- package/src/server/index/serve-options.ts +5 -4
- package/src/server/local-management-read-client.ts +3 -1
- package/src/server/management/agent-settings-routes.ts +27 -15
- package/src/server/management/combo-routes.ts +14 -4
- package/src/server/management/config-routes.ts +2 -2
- package/src/server/management/decision-discovery.ts +12 -2
- package/src/server/management/hermes-profile-routes.ts +156 -0
- package/src/server/management/integration-routes.ts +85 -24
- package/src/server/management/link-routes.ts +5 -2
- package/src/server/management/model-routes.ts +2 -2
- package/src/server/management/model-row-export-metadata.ts +25 -6
- package/src/server/management/model-rows.ts +15 -20
- package/src/server/management/native-integration-routes.ts +11 -8
- package/src/server/management/provider-routes.ts +1 -1
- package/src/server/management/route-registry.ts +4 -0
- package/src/server/management-api.ts +5 -0
- package/src/server/models-capabilities.ts +21 -3
- package/src/server/port-reclaim.ts +51 -9
- package/src/server/request-log-conversation.ts +49 -0
- package/src/server/request-log.ts +2 -0
- package/src/server/responses/adapter-delivery.ts +11 -1
- package/src/server/responses/adapter-dispatch.ts +2 -2
- package/src/server/responses/anthropic-thinking-recovery.ts +1 -0
- package/src/server/responses/codex-ws-exchange.ts +39 -9
- package/src/server/responses/codex-ws-session.ts +38 -4
- package/src/server/responses/codex-ws-wire.ts +6 -1
- package/src/server/responses/combo-required-tool-contract.ts +103 -0
- package/src/server/responses/combo-stream-preflight.ts +8 -1
- package/src/server/responses/compact.ts +14 -5
- package/src/server/responses/context-overflow.ts +8 -1
- package/src/server/responses/core-combo-failure.ts +10 -0
- package/src/server/responses/core-combo.ts +74 -44
- package/src/server/responses/core-lifetime.ts +5 -2
- package/src/server/responses/core-normalize.ts +17 -4
- package/src/server/responses/core-opaque-recovery.ts +1 -0
- package/src/server/responses/core-options.ts +7 -0
- package/src/server/responses/core-replay.ts +2 -1
- package/src/server/responses/jev-combo-choice.ts +39 -0
- package/src/server/responses/jev-session-pin.ts +203 -0
- package/src/server/responses/main-hard-lock-routing.ts +60 -44
- package/src/server/responses/passthrough-delivery.ts +44 -2
- package/src/server/responses/policy-fallback.ts +6 -1
- package/src/server/responses/request-compression.ts +12 -7
- package/src/server/responses/request-prepare.ts +32 -7
- package/src/server/responses/request-sidecar-auth.ts +2 -0
- package/src/server/responses/request-transport.ts +25 -6
- package/src/server/responses/response-effects.ts +22 -10
- package/src/server/responses/run-turn-execution.ts +2 -4
- package/src/server/responses/sidecar-execution.ts +13 -7
- package/src/server/responses-phase-inference.ts +267 -0
- package/src/server/sse-payload-rewrite.ts +14 -1
- package/src/server/stop-teardown.ts +36 -1
- package/src/server/ws-bridge.ts +1 -1
- package/src/service/definition.ts +20 -8
- package/src/service/desktop-supervision.d.mts +7 -4
- package/src/service/desktop-supervision.mjs +90 -13
- package/src/service/ownership-mutation-lease.mjs +21 -7
- package/src/types/config.ts +14 -2
- package/src/types/provider.ts +29 -7
- package/src/types/request.ts +11 -0
- package/src/types/wire.ts +1 -0
- package/src/types.ts +1 -0
- package/src/update/bind-probe.ts +29 -0
- package/src/update/codex-cli-update-launch-policy.d.mts +9 -0
- package/src/update/codex-cli-update-launch-policy.mjs +30 -0
- package/src/update/job.ts +10 -29
- package/src/update/npm-install-job.ps1 +173 -0
- package/src/update/npm-install.mjs +70 -0
- package/src/update/stop-contract.d.mts +1 -0
- package/src/update/stop-contract.mjs +14 -0
- package/src/update/stop-decision.d.mts +1 -1
- package/src/update/stop-decision.mjs +13 -3
- package/src/update/teardown-handover.ts +166 -0
- package/src/update/transactional-install.d.mts +14 -3
- package/src/update/transactional-install.mjs +33 -7
- package/src/usage/expected-prices.ts +64 -31
- package/src/usage/jev-stats.ts +54 -14
- package/src/usage/timeline.ts +19 -9
- package/src/vision/eligibility.ts +15 -5
- package/src/vision/plan.ts +2 -2
- package/src/web-search/loop.ts +16 -2
- package/gui/dist/assets/App-nZ34cR5G.js +0 -58
- package/gui/dist/assets/index-DGXS51sp.js +0 -94
- package/gui/dist/assets/index-DPeB9HpV.css +0 -1
- package/gui/dist/assets/quota-summary-DO1Sx5TM.js +0 -1
|
@@ -155,30 +155,33 @@ export const ZAI_GLM_5X_REASONING_EFFORTS: Record<string, string[]> = {
|
|
|
155
155
|
};
|
|
156
156
|
// 260710 MiniMax models and context windows: Tier-2 evidence in
|
|
157
157
|
// devlog/_plan/260710_provider_hardening/002_research_cn.md.
|
|
158
|
-
//
|
|
159
|
-
//
|
|
160
|
-
//
|
|
161
|
-
|
|
162
|
-
|
|
158
|
+
// 261021 MiniMax-M3.1-Flash: current global/CN text route, 1M context, thinking always on
|
|
159
|
+
// (effort none or thinking disabled answers 400), omitted effort = max. It returns thinking as
|
|
160
|
+
// reasoning_content and ignores reasoning_split. The live /v1/models roster may omit it.
|
|
161
|
+
export const MINIMAX_M31_FLASH = "MiniMax-M3.1-Flash";
|
|
162
|
+
/** The preview id MiniMax retired when M3.1-Flash went GA (2026-10-19); see MODEL_RENAMES. */
|
|
163
|
+
export const MINIMAX_M31_FLASH_PREVIEW_RETIRED = "MiniMax-M3.1-Flash-Preview";
|
|
164
|
+
/** MiniMax's published M3.1-Flash output ceiling (512K). */
|
|
165
|
+
export const MINIMAX_M31_FLASH_MAX_OUTPUT_TOKENS = 524_288;
|
|
163
166
|
export const MINIMAX_MODELS = [
|
|
164
|
-
|
|
167
|
+
MINIMAX_M31_FLASH,
|
|
165
168
|
"MiniMax-M3",
|
|
166
169
|
"MiniMax-M2.7", "MiniMax-M2.7-highspeed",
|
|
167
170
|
"MiniMax-M2.5", "MiniMax-M2.5-highspeed",
|
|
168
171
|
"MiniMax-M2.1", "MiniMax-M2.1-highspeed",
|
|
169
172
|
"MiniMax-M2",
|
|
170
173
|
];
|
|
171
|
-
/** The eight-id roster every MiniMax preset seeded
|
|
172
|
-
export const MINIMAX_MODELS_BEFORE_M31 = MINIMAX_MODELS.filter(id => id !==
|
|
174
|
+
/** The eight-id roster every MiniMax preset seeded before Flash landed. */
|
|
175
|
+
export const MINIMAX_MODELS_BEFORE_M31 = MINIMAX_MODELS.filter(id => id !== MINIMAX_M31_FLASH);
|
|
173
176
|
/** Models that honour reasoning_split and answer with structured reasoning_details. */
|
|
174
177
|
export const MINIMAX_REASONING_SPLIT_MODELS = MINIMAX_MODELS_BEFORE_M31;
|
|
175
178
|
export const MINIMAX_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
|
|
176
|
-
MINIMAX_MODELS.map(id => [id, id === "MiniMax-M3" || id ===
|
|
179
|
+
MINIMAX_MODELS.map(id => [id, id === "MiniMax-M3" || id === MINIMAX_M31_FLASH ? 1_000_000 : 204_800]),
|
|
177
180
|
);
|
|
178
181
|
/** MiniMax's OpenAI-compatible M3 endpoints accept image_url; keep video off Codex's input enum. */
|
|
179
182
|
export const MINIMAX_MODEL_INPUT_MODALITIES: Record<string, string[]> = {
|
|
180
183
|
"MiniMax-M3": ["text", "image"],
|
|
181
|
-
[
|
|
184
|
+
[MINIMAX_M31_FLASH]: ["text", "image"],
|
|
182
185
|
};
|
|
183
186
|
export const MINIMAX_M3_REASONING_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
|
|
184
187
|
/** Identity efforts on the wire; no map, so none omits the field instead of disabling thinking. */
|
|
@@ -219,7 +222,8 @@ export const OPENAI_API_GPT56_REASONING_EFFORTS = ["low", "medium", "high", "xhi
|
|
|
219
222
|
* 1,050,000 context, 922,000 max input, 128,000 max output, efforts low..max
|
|
220
223
|
* (https://developers.openai.com/api/docs/models/gpt-6.1-sol, checked 2026-09-30).
|
|
221
224
|
*/
|
|
222
|
-
|
|
225
|
+
// GPT-6.1 Terra (2026-10-26): same API limits; its none..max ladder is set per model.
|
|
226
|
+
export const OPENAI_GPT6_MODELS = ["gpt-6.1-sol", "gpt-6.1-terra", "gpt-6-sol", "gpt-6-luna"];
|
|
223
227
|
/*
|
|
224
228
|
* Meta Model API (https://api.meta.ai/v1) — published ladder, deliberately NOT the
|
|
225
229
|
* house set. dev.meta.ai/docs/reasoning lists "none", "minimal", "low", "medium",
|
|
@@ -327,6 +331,8 @@ export const OPENROUTER_GPT56_CONTEXT_WINDOWS = {
|
|
|
327
331
|
"openai/gpt-6-luna": OPENROUTER_GPT56_CONTEXT_WINDOW,
|
|
328
332
|
// Live /api/v1/models on 2026-09-30: context_length 1,050,000, max_completion_tokens 128,000.
|
|
329
333
|
"openai/gpt-6.1-sol": OPENROUTER_GPT56_CONTEXT_WINDOW,
|
|
334
|
+
// OpenRouter listing checked 2026-10-27: context_length 1,050,000.
|
|
335
|
+
"openai/gpt-6.1-terra": OPENROUTER_GPT56_CONTEXT_WINDOW,
|
|
330
336
|
};
|
|
331
337
|
|
|
332
338
|
/**
|
|
@@ -377,8 +383,8 @@ export const OPENCODE_GO_THINKING_BUDGET_MODELS = ["qwen3.5-plus", "qwen3.6-plus
|
|
|
377
383
|
/*
|
|
378
384
|
* DeepSeek moved the whole V4 name set on 2026-09-10. V4.1-Flash ships as deepseek-flash
|
|
379
385
|
* on the first-party API; deepseek-v4-flash and the vision preview retire as models but
|
|
380
|
-
* keep routing there as compatibility aliases
|
|
381
|
-
*
|
|
386
|
+
* keep routing there as compatibility aliases. V4 Pro remains available: the planned
|
|
387
|
+
* retirement was withdrawn. Evidence: https://api-docs.deepseek.com/updates/ (2026-10-23).
|
|
382
388
|
*
|
|
383
389
|
* The spelling differs by who serves it, so one shared list cannot express it: the
|
|
384
390
|
* first-party API answers to deepseek-flash, while the Zen gateway exposes the route as
|
|
@@ -388,15 +394,10 @@ export const OPENCODE_GO_THINKING_BUDGET_MODELS = ["qwen3.5-plus", "qwen3.6-plus
|
|
|
388
394
|
*/
|
|
389
395
|
export const DEEPSEEK_V4_LEGACY_MODELS = ["deepseek-v4-flash"];
|
|
390
396
|
/*
|
|
391
|
-
*
|
|
392
|
-
*
|
|
393
|
-
* row here would advertise a Pro context window and Pro pricing for a route that serves
|
|
394
|
-
* Flash. The retirement is followed through every roster in this file, including the
|
|
395
|
-
* vendor-hosted ones; providers that discover their models live are handled by
|
|
396
|
-
* `ROUTED_MODEL_COMPATIBILITY_EXCLUSIONS` because deleting a row there removes the
|
|
397
|
-
* model's capabilities rather than the model.
|
|
397
|
+
* The official API continues serving V4 Pro after September 14 with unchanged billing.
|
|
398
|
+
* Gateway availability is independent; restoring the native set does not change gateway rosters.
|
|
398
399
|
*/
|
|
399
|
-
export const DEEPSEEK_NATIVE_THINKING_MODELS = ["deepseek-flash", "deepseek-v4-flash"];
|
|
400
|
+
export const DEEPSEEK_NATIVE_THINKING_MODELS = ["deepseek-flash", "deepseek-v4-flash", "deepseek-v4-pro"];
|
|
400
401
|
export const DEEPSEEK_GATEWAY_THINKING_MODELS = ["deepseek-v4.1-flash", "deepseek-v4-flash"];
|
|
401
402
|
/*
|
|
402
403
|
* DeepSeek's legacy vision preview id (released 2026-08-21). First-party probes
|
|
@@ -573,22 +574,22 @@ export const deepseekReasoningMapFor = (modelId: string): Record<string, string>
|
|
|
573
574
|
// text-only input, strict json_schema accepted. glm-5.3-flash REMAINS OUT:
|
|
574
575
|
// still never served by the Token Plan gateway (docs.z.ai VLM id, not plan
|
|
575
576
|
// entitlement).
|
|
576
|
-
//
|
|
577
|
-
//
|
|
578
|
-
//
|
|
579
|
-
//
|
|
580
|
-
//
|
|
581
|
-
//
|
|
582
|
-
//
|
|
583
|
-
//
|
|
584
|
-
//
|
|
585
|
-
//
|
|
586
|
-
//
|
|
587
|
-
//
|
|
588
|
-
//
|
|
577
|
+
// Both regions support Personal and Team subscriptions. Keep the existing Beijing
|
|
578
|
+
// Personal and International Team model rosters. Discovery stays disabled because
|
|
579
|
+
// the upstream list includes non-chat models.
|
|
580
|
+
// 261020: 0813 is now listed on Global Personal/Team and CN Team, and the CN Personal
|
|
581
|
+
// overview documents entitlement. Add it to the existing Personal curation without
|
|
582
|
+
// claiming new Personal inference. Retired 261011: deepseek-v3.2 and bare v4-flash.
|
|
583
|
+
// 1017 evidence: https://www.qianwenai.com/models/deepseek-v4.1-flash-1017 and
|
|
584
|
+
// https://docs.qwencloud.com/api-reference/chat/openai-chat (checked 2026-10-20).
|
|
585
|
+
// Global Personal probes: completion, thinking-off, low effort, tools, user/tool images
|
|
586
|
+
// and Responses succeed; strict json_schema rejects 400. json_object and replay are
|
|
587
|
+
// untested on this exact id. Published 1M context and 393K output use 393216 here,
|
|
588
|
+
// consistent with the existing plan budget convention, not a measured boundary;
|
|
589
|
+
// the lab's 384000 precision differs. Do not inherit the bare model's JSON fallback.
|
|
589
590
|
export const ALIBABA_TOKEN_PLAN_MODELS = [
|
|
590
591
|
"qwen3.8-max", "qwen3.8-flash", "qwen3.7-max", "qwen3.7-plus", "qwen3.6-flash",
|
|
591
|
-
"deepseek-v4-pro", "deepseek-v4-flash-0731", "deepseek-v4.1-flash", "glm-5.2", "glm-5.3",
|
|
592
|
+
"deepseek-v4-pro", "deepseek-v4-pro-0813", "deepseek-v4-flash-0731", "deepseek-v4.1-flash", "deepseek-v4.1-flash-1017", "glm-5.2", "glm-5.3",
|
|
592
593
|
];
|
|
593
594
|
export const ALIBABA_TOKEN_PLAN_QWEN_MODELS = [
|
|
594
595
|
"qwen3.8-max", "qwen3.8-flash", "qwen3.7-max", "qwen3.7-plus", "qwen3.6-flash",
|
|
@@ -604,12 +605,13 @@ export const ALIBABA_TOKEN_PLAN_INPUT_MODALITIES: Record<string, string[]> = {
|
|
|
604
605
|
"deepseek-v4-flash-0731": ["text"],
|
|
605
606
|
// Vision probed on the plan gateway 260915 (user message and tool result, both 200).
|
|
606
607
|
"deepseek-v4.1-flash": ["text", "image"],
|
|
608
|
+
"deepseek-v4.1-flash-1017": ["text", "image"],
|
|
607
609
|
"glm-5.2": ["text"],
|
|
608
610
|
"glm-5.3": ["text"],
|
|
609
611
|
};
|
|
610
612
|
|
|
611
613
|
// 260721 Alibaba Token Plan International (ap-southeast-1 / Singapore, hardened 260721).
|
|
612
|
-
//
|
|
614
|
+
// International Team model roster; tier and region are independent.
|
|
613
615
|
// Evidence: https://www.alibabacloud.com/help/en/model-studio/token-plan-overview
|
|
614
616
|
// https://qwencloud.com/pricing/token-plan (qwen3.8 metadata)
|
|
615
617
|
// The Team Edition roster (Singapore), verified identical to the CN Team set on 260909.
|
|
@@ -617,7 +619,7 @@ export const ALIBABA_TOKEN_PLAN_INPUT_MODALITIES: Record<string, string[]> = {
|
|
|
617
619
|
// listed on /models on both regions) after being dropped as "retired" upstream.
|
|
618
620
|
export const ALIBABA_INTL_TOKEN_PLAN_MODELS = [
|
|
619
621
|
"qwen3.8-max", "qwen3.8-flash", "qwen3.7-max", "qwen3.7-plus", "qwen3.6-plus", "qwen3.6-flash",
|
|
620
|
-
"deepseek-v4-pro", "deepseek-v4-pro-0813", "deepseek-v4-flash", "deepseek-v4-flash
|
|
622
|
+
"deepseek-v4-pro", "deepseek-v4-pro-0813", "deepseek-v4-flash-0731", "deepseek-v4.1-flash", "deepseek-v4.1-flash-1017",
|
|
621
623
|
"kimi-k2.7-code", "kimi-k2.6", "kimi-k2.5",
|
|
622
624
|
"glm-5.2", "glm-5.3", "glm-5.1", "glm-5",
|
|
623
625
|
"MiniMax-M2.5",
|
|
@@ -702,8 +704,6 @@ export const VOLCENGINE_PLAN_TEXT_ONLY_MODELS = [
|
|
|
702
704
|
export const ALIBABA_INTL_TOKEN_PLAN_INPUT_MODALITIES: Record<string, string[]> = {
|
|
703
705
|
...ALIBABA_TOKEN_PLAN_INPUT_MODALITIES,
|
|
704
706
|
"qwen3.6-plus": ["text", "image"],
|
|
705
|
-
"deepseek-v4-flash": ["text"],
|
|
706
|
-
"deepseek-v3.2": ["text"],
|
|
707
707
|
"kimi-k2.7-code": ["text", "image"],
|
|
708
708
|
"kimi-k2.6": ["text", "image"],
|
|
709
709
|
"kimi-k2.5": ["text", "image"],
|
|
@@ -712,14 +712,14 @@ export const ALIBABA_INTL_TOKEN_PLAN_INPUT_MODALITIES: Record<string, string[]>
|
|
|
712
712
|
"MiniMax-M2.5": ["text"],
|
|
713
713
|
};
|
|
714
714
|
|
|
715
|
-
// Shared Token Plan metadata
|
|
716
|
-
//
|
|
715
|
+
// Shared Token Plan metadata: legacy rows use 260909 max_tokens boundary probes.
|
|
716
|
+
// The 1017 row uses published card normalization, not a boundary probe (see above).
|
|
717
717
|
export const QWEN38_FAMILY = ["qwen3.8-max", "qwen3.8-flash"];
|
|
718
718
|
export const ALIBABA_TOKEN_PLAN_CONTEXT_WINDOWS: Record<string, number> = {
|
|
719
719
|
"qwen3.8-max": 1_000_000, "qwen3.8-flash": 1_000_000, "qwen3.7-max": 1_000_000, "qwen3.7-plus": 1_000_000,
|
|
720
720
|
"qwen3.6-plus": 1_000_000, "qwen3.6-flash": 1_000_000,
|
|
721
|
-
"deepseek-v4-pro": 1_000_000, "deepseek-v4-pro-0813": 1_000_000,
|
|
722
|
-
"deepseek-v4-flash-0731": 1_000_000, "deepseek-v4.1-flash": 1_000_000, "deepseek-
|
|
721
|
+
"deepseek-v4-pro": 1_000_000, "deepseek-v4-pro-0813": 1_000_000,
|
|
722
|
+
"deepseek-v4-flash-0731": 1_000_000, "deepseek-v4.1-flash": 1_000_000, "deepseek-v4.1-flash-1017": 1_000_000,
|
|
723
723
|
"kimi-k2.7-code": 262_144, "kimi-k2.6": 262_144, "kimi-k2.5": 262_144,
|
|
724
724
|
"glm-5.2": 1_000_000, "glm-5.3": 1_000_000, "glm-5.1": 202_752, "glm-5": 202_752,
|
|
725
725
|
"MiniMax-M2.5": 196_608,
|
|
@@ -727,20 +727,20 @@ export const ALIBABA_TOKEN_PLAN_CONTEXT_WINDOWS: Record<string, number> = {
|
|
|
727
727
|
export const ALIBABA_TOKEN_PLAN_MAX_OUTPUT_TOKENS: Record<string, number> = {
|
|
728
728
|
"qwen3.8-max": 131_072, "qwen3.8-flash": 131_072, "qwen3.7-max": 131_072, "qwen3.7-plus": 131_072,
|
|
729
729
|
"qwen3.6-plus": 65_536, "qwen3.6-flash": 65_536,
|
|
730
|
-
"deepseek-v4-pro": 393_216, "deepseek-v4-pro-0813": 393_216,
|
|
731
|
-
"deepseek-v4-flash-0731": 393_216, "deepseek-v4.1-flash": 393_216, "deepseek-
|
|
730
|
+
"deepseek-v4-pro": 393_216, "deepseek-v4-pro-0813": 393_216,
|
|
731
|
+
"deepseek-v4-flash-0731": 393_216, "deepseek-v4.1-flash": 393_216, "deepseek-v4.1-flash-1017": 393_216,
|
|
732
732
|
"kimi-k2.7-code": 262_144, "kimi-k2.6": 262_144, "kimi-k2.5": 98_304,
|
|
733
733
|
"glm-5.2": 131_072, "glm-5.3": 131_072, "glm-5.1": 128_000, "glm-5": 16_384,
|
|
734
734
|
"MiniMax-M2.5": 32_768,
|
|
735
735
|
};
|
|
736
736
|
export const ALIBABA_TOKEN_PLAN_NO_VISION = [
|
|
737
|
-
"qwen3.7-max", "deepseek-v4-pro", "deepseek-v4-pro-0813",
|
|
738
|
-
"deepseek-v4-flash-0731", "
|
|
737
|
+
"qwen3.7-max", "deepseek-v4-pro", "deepseek-v4-pro-0813",
|
|
738
|
+
"deepseek-v4-flash-0731", "glm-5.2", "glm-5.3", "glm-5.1", "glm-5", "MiniMax-M2.5",
|
|
739
739
|
];
|
|
740
740
|
export const ALIBABA_TOKEN_PLAN_PRESERVE_REASONING = [
|
|
741
741
|
"qwen3.8-max", "qwen3.8-flash", "qwen3.7-max", "qwen3.7-plus", "qwen3.6-plus", "qwen3.6-flash",
|
|
742
|
-
"deepseek-v4-pro", "deepseek-v4-pro-0813", "deepseek-v4-flash
|
|
743
|
-
"deepseek-v4.1-flash", "glm-5.2", "glm-5.3",
|
|
742
|
+
"deepseek-v4-pro", "deepseek-v4-pro-0813", "deepseek-v4-flash-0731",
|
|
743
|
+
"deepseek-v4.1-flash", "deepseek-v4.1-flash-1017", "glm-5.2", "glm-5.3",
|
|
744
744
|
];
|
|
745
745
|
|
|
746
746
|
// 260717 Kimi K3: the subscription endpoint uses one upstream id (`k3`) for both
|
|
@@ -754,6 +754,12 @@ export const ALIBABA_TOKEN_PLAN_PRESERVE_REASONING = [
|
|
|
754
754
|
export const KIMI_K3_STANDARD_CONTEXT_WINDOW = 262_144;
|
|
755
755
|
export const KIMI_K3_1M_CONTEXT_WINDOW = 1_048_576;
|
|
756
756
|
export const KIMI_CODING_K3_MODELS = ["k3", "k3[1m]", "k3-256k"];
|
|
757
|
+
// 261024 Kimi K3.1 Pro: live GET /coding/v1/models now also lists k3.1-pro and
|
|
758
|
+
// k3.1-pro-highspeed, both advertising context_length 1048576 (flat 1M, no entitlement
|
|
759
|
+
// alias). Both run thinking-only: the endpoint rejects thinking-disabled requests with
|
|
760
|
+
// HTTP 400, and an absent effort defaults to max — the same low/high/max ladder as k3.
|
|
761
|
+
// Evidence: https://www.kimi.com/code/docs/en/kimi-code/models.html
|
|
762
|
+
export const KIMI_CODING_K31_MODELS = ["k3.1-pro", "k3.1-pro-highspeed"];
|
|
757
763
|
// 260921 Kimi K2.8: `kimi-for-coding` is the stable subscription alias Moonshot re-points
|
|
758
764
|
// at each coding release. Live GET /coding/v1/models lists only kimi-for-coding[-highspeed],
|
|
759
765
|
// k3, k3-256k — the k2.x ids are retired from the subscription endpoint. Since K2.8 Preview
|
|
@@ -762,10 +768,27 @@ export const KIMI_CODING_K3_MODELS = ["k3", "k3[1m]", "k3-256k"];
|
|
|
762
768
|
// with "model token limit: 1048576" beyond that.
|
|
763
769
|
// Evidence: https://www.kimi.com/code/docs/en/kimi-code/models.html
|
|
764
770
|
export const KIMI_CODING_K28_MODELS = ["kimi-for-coding"];
|
|
765
|
-
export const KIMI_CODING_LIVE_MODELS = [...KIMI_CODING_K3_MODELS, ...KIMI_CODING_K28_MODELS];
|
|
771
|
+
export const KIMI_CODING_LIVE_MODELS = [...KIMI_CODING_K3_MODELS, ...KIMI_CODING_K31_MODELS, ...KIMI_CODING_K28_MODELS];
|
|
766
772
|
export const KIMI_CODING_ADJUSTABLE_THINKING_MODELS = [...KIMI_CODING_LIVE_MODELS];
|
|
767
773
|
export const KIMI_LEGACY_API_MODELS = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
|
|
768
|
-
|
|
774
|
+
// 261025 Kimi K3.1 Pro on the Moonshot API (#7653): the platform model list names kimi-k3.1-pro
|
|
775
|
+
// and kimi-k3.1-pro-highspeed (same model, 2-3x speed) on both the International and China
|
|
776
|
+
// platforms. Every K3-series model has a 1,048,576-token window, and both ids share kimi-k3's
|
|
777
|
+
// parameter constraints: thinking is always on, top-level reasoning_effort takes low/high/max
|
|
778
|
+
// (default max), temperature/top_p/n/penalties are fixed, tool_choice "required" is supported.
|
|
779
|
+
// Verified 2026-10-25 against official docs only:
|
|
780
|
+
// Evidence: https://platform.kimi.ai/docs/models.md
|
|
781
|
+
// https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart.md
|
|
782
|
+
// https://platform.kimi.ai/docs/api/models-overview.md
|
|
783
|
+
// https://platform.kimi.ai/docs/pricing/chat.md
|
|
784
|
+
// https://platform.kimi.com/docs/models.md (China)
|
|
785
|
+
export const KIMI_API_K31_MODELS = ["kimi-k3.1-pro", "kimi-k3.1-pro-highspeed"];
|
|
786
|
+
export const KIMI_API_K3_SERIES_MODELS = ["kimi-k3", ...KIMI_API_K31_MODELS];
|
|
787
|
+
export const KIMI_API_MODELS = [...KIMI_API_K3_SERIES_MODELS, ...KIMI_LEGACY_API_MODELS];
|
|
788
|
+
// The moonshot roster saved by installs between the K3 seed (260717) and K3.1 Pro. A literal on
|
|
789
|
+
// purpose: stale-model-roster-migration refreshes only rows still identical to it, and a later
|
|
790
|
+
// K2 retirement must not change what that refresh matches.
|
|
791
|
+
export const KIMI_API_MODELS_BEFORE_K31 = ["kimi-k3", "kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
|
|
769
792
|
// Every kimi coding preset record - picker, context windows, locked-parameter lists -
|
|
770
793
|
// derives from the live ids only. seeding a retired id in a metadata list would re-arm
|
|
771
794
|
// the model-rename migration on every boot (#5066): the list holds the retired id but
|
|
@@ -774,7 +797,7 @@ export const KIMI_API_MODELS = ["kimi-k3", ...KIMI_LEGACY_API_MODELS];
|
|
|
774
797
|
// repairs saved rows still naming them.
|
|
775
798
|
export const KIMI_THINKING_MODELS = KIMI_CODING_LIVE_MODELS;
|
|
776
799
|
export const KIMI_CODING_NO_REASONING_MODELS = KIMI_CODING_LIVE_MODELS.filter(id => !KIMI_CODING_ADJUSTABLE_THINKING_MODELS.includes(id));
|
|
777
|
-
export const KIMI_API_NO_REASONING_MODELS = KIMI_API_MODELS.filter(id => id
|
|
800
|
+
export const KIMI_API_NO_REASONING_MODELS = KIMI_API_MODELS.filter(id => !KIMI_API_K3_SERIES_MODELS.includes(id));
|
|
778
801
|
export const KIMI_CODING_K3_REASONING_EFFORTS = ["low", "high", "max"];
|
|
779
802
|
export const KIMI_CODING_K3_REASONING_EFFORT_MAP: Record<string, string> = {
|
|
780
803
|
none: "none",
|
|
@@ -793,15 +816,33 @@ export const KIMI_CODING_DEFAULT_REASONING_EFFORTS = Object.fromEntries(
|
|
|
793
816
|
export const KIMI_CODING_REASONING_EFFORT_MAPS = Object.fromEntries(
|
|
794
817
|
KIMI_CODING_ADJUSTABLE_THINKING_MODELS.map(id => [id, KIMI_CODING_K3_REASONING_EFFORT_MAP]),
|
|
795
818
|
);
|
|
819
|
+
// kimi-k3 keeps its launch-time max-only list; K3.1 Pro gets the documented low/high/max ladder.
|
|
796
820
|
export const KIMI_API_REASONING_EFFORTS = Object.fromEntries(
|
|
797
|
-
KIMI_API_MODELS.map(id => [
|
|
821
|
+
KIMI_API_MODELS.map(id => [
|
|
822
|
+
id,
|
|
823
|
+
KIMI_API_K31_MODELS.includes(id) ? KIMI_CODING_K3_REASONING_EFFORTS : id === "kimi-k3" ? ["max"] : [],
|
|
824
|
+
]),
|
|
825
|
+
);
|
|
826
|
+
export const KIMI_API_DEFAULT_REASONING_EFFORTS = Object.fromEntries(KIMI_API_K31_MODELS.map(id => [id, "max"]));
|
|
827
|
+
// K3.1 Pro cannot disable thinking and documents only low/high/max, so an explicit "none" omits
|
|
828
|
+
// reasoning_effort (upstream default max) rather than sending an undocumented value. The value is
|
|
829
|
+
// REASONING_EFFORT_OMIT_SENTINEL (src/reasoning-effort.ts); the registry does not import the
|
|
830
|
+
// request-path module, and provider-registry-parity asserts the two stay equal.
|
|
831
|
+
export const KIMI_API_K31_REASONING_EFFORT_MAP: Record<string, string> = {
|
|
832
|
+
...KIMI_CODING_K3_REASONING_EFFORT_MAP,
|
|
833
|
+
none: "__omit__",
|
|
834
|
+
};
|
|
835
|
+
export const KIMI_API_REASONING_EFFORT_MAPS = Object.fromEntries(
|
|
836
|
+
KIMI_API_K31_MODELS.map(id => [id, KIMI_API_K31_REASONING_EFFORT_MAP]),
|
|
798
837
|
);
|
|
799
838
|
export const KIMI_LOCKED_PARAMETER_MODELS = KIMI_CODING_LIVE_MODELS;
|
|
800
839
|
export const KIMI_AUTO_TOOL_CHOICE_ONLY_MODELS = ["kimi-for-coding"];
|
|
801
840
|
export const KIMI_API_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
|
|
802
|
-
KIMI_API_MODELS.map(id => [id, id
|
|
841
|
+
KIMI_API_MODELS.map(id => [id, KIMI_API_K3_SERIES_MODELS.includes(id) ? KIMI_K3_1M_CONTEXT_WINDOW : 262_144]),
|
|
842
|
+
);
|
|
843
|
+
export const KIMI_API_MODEL_INPUT_MODALITIES: Record<string, string[]> = Object.fromEntries(
|
|
844
|
+
KIMI_API_K3_SERIES_MODELS.map(id => [id, ["text", "image"]]),
|
|
803
845
|
);
|
|
804
|
-
export const KIMI_API_MODEL_INPUT_MODALITIES = { "kimi-k3": ["text", "image"] };
|
|
805
846
|
|
|
806
847
|
// 260715 NVIDIA NIM kimi family (issue #126): documented served ids on integrate
|
|
807
848
|
// chat/completions per docs.api.nvidia.com/nim/reference/llm-apis; live /v1/models
|
|
@@ -885,7 +926,7 @@ export const NVIDIA_NIM_NO_VISION_MODELS = [
|
|
|
885
926
|
"poolside/laguna-xs-2.1", "z-ai/glm-5.3", "z-ai/glm-5.2",
|
|
886
927
|
];
|
|
887
928
|
export const KIMI_CODING_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
|
|
888
|
-
KIMI_CODING_LIVE_MODELS.map(id => [id, (id === "k3[1m]" || KIMI_CODING_K28_MODELS.includes(id)) ? KIMI_K3_1M_CONTEXT_WINDOW : KIMI_K3_STANDARD_CONTEXT_WINDOW]),
|
|
929
|
+
KIMI_CODING_LIVE_MODELS.map(id => [id, (id === "k3[1m]" || KIMI_CODING_K31_MODELS.includes(id) || KIMI_CODING_K28_MODELS.includes(id)) ? KIMI_K3_1M_CONTEXT_WINDOW : KIMI_K3_STANDARD_CONTEXT_WINDOW]),
|
|
889
930
|
);
|
|
890
931
|
export const KIMI_CODING_MODEL_INPUT_MODALITIES = Object.fromEntries(
|
|
891
932
|
KIMI_CODING_ADJUSTABLE_THINKING_MODELS.map(id => [id, ["text", "image"]]),
|
|
@@ -949,6 +990,8 @@ export const DIGITALOCEAN_CHAT_COMPLETION_MODELS = [
|
|
|
949
990
|
"arcee-trinity-large-thinking",
|
|
950
991
|
"openai-gpt-5.6-sol",
|
|
951
992
|
"openai-gpt-5.6-terra",
|
|
993
|
+
// 261027 preemptive: GPT-6.1 Terra (mirrors the GPT-6 Sol seed; provider listing unverified).
|
|
994
|
+
"openai-gpt-6.1-terra",
|
|
952
995
|
"openai-gpt-5.6-luna",
|
|
953
996
|
// 260923 preemptive: GPT-6 Sol/Luna ahead of DigitalOcean's model list.
|
|
954
997
|
"openai-gpt-6-sol",
|
|
@@ -1166,6 +1209,27 @@ export const OPPER_MODEL_INPUT_MODALITIES: Record<string, string[]> = Object.fro
|
|
|
1166
1209
|
OPPER_MODELS.map(id => [id, OPPER_TEXT_ONLY_MODELS.includes(id) ? ["text"] : ["text", "image"]]),
|
|
1167
1210
|
);
|
|
1168
1211
|
|
|
1212
|
+
// AI Stupid Level routing modes, as listed by the authenticated GET /v1/models on 2026-10-26
|
|
1213
|
+
// (https://aistupidlevel.info/router/docs#strategies). Each mode picks a model per request, so it
|
|
1214
|
+
// declares conservative limits that hold for any pick: a 262,144-token window (the pool's smallest,
|
|
1215
|
+
// Kimi K2.7 Code's, when first checked on 2026-10-23) and 64,000 output tokens.
|
|
1216
|
+
// The router is text only (image, audio and file parts are a 400 `unsupported_content`), so images
|
|
1217
|
+
// for a mode go through the vision sidecar. Pinned model ids arrive through live discovery.
|
|
1218
|
+
export const AISTUPIDLEVEL_ROUTING_MODES = [
|
|
1219
|
+
"auto-coding", "auto-tooling", "auto-reasoning", "auto-best", "auto", "auto-task",
|
|
1220
|
+
"auto-value", "auto-value-coding", "auto-value-tooling", "auto-value-reasoning",
|
|
1221
|
+
"auto-fastest-quality", "auto-consistent", "auto-cheapest", "auto-fastest", "auto-split",
|
|
1222
|
+
];
|
|
1223
|
+
export const AISTUPIDLEVEL_MODEL_CONTEXT_WINDOWS: Record<string, number> = Object.fromEntries(
|
|
1224
|
+
AISTUPIDLEVEL_ROUTING_MODES.map(id => [id, 262_144]),
|
|
1225
|
+
);
|
|
1226
|
+
export const AISTUPIDLEVEL_MODEL_MAX_OUTPUT_TOKENS: Record<string, number> = Object.fromEntries(
|
|
1227
|
+
AISTUPIDLEVEL_ROUTING_MODES.map(id => [id, 64_000]),
|
|
1228
|
+
);
|
|
1229
|
+
export const AISTUPIDLEVEL_MODEL_INPUT_MODALITIES: Record<string, string[]> = Object.fromEntries(
|
|
1230
|
+
AISTUPIDLEVEL_ROUTING_MODES.map(id => [id, ["text"]]),
|
|
1231
|
+
);
|
|
1232
|
+
|
|
1169
1233
|
export const STEPFUN_MODELS = [
|
|
1170
1234
|
"step-5-preview",
|
|
1171
1235
|
"step-3.5-flash",
|
|
@@ -63,6 +63,8 @@ export interface ProviderModelDiscoveryFilter {
|
|
|
63
63
|
}
|
|
64
64
|
|
|
65
65
|
interface ProviderModelDiscoverySharedSpec {
|
|
66
|
+
/** Product-config envelope used by WorkBuddy AI instead of an OpenAI model list. */
|
|
67
|
+
format?: "workbuddy";
|
|
66
68
|
/** Query parameters applied to the resolved discovery URL. */
|
|
67
69
|
query?: Readonly<Record<string, string>>;
|
|
68
70
|
/** Top-level response key containing model rows; defaults to `data`. */
|
|
@@ -160,6 +162,14 @@ export interface ProviderRegistryEntry {
|
|
|
160
162
|
*/
|
|
161
163
|
sponsor?: { tier: "main" | "standard"; url: string };
|
|
162
164
|
dashboardPreset?: boolean;
|
|
165
|
+
/**
|
|
166
|
+
* The upstream route or service this entry depends on is gone. Deprecated entries stay in the
|
|
167
|
+
* registry so saved providers keep resolving and a pinned adapter keeps loading, but they are
|
|
168
|
+
* excluded from the add-provider picker, the init menu, the featured list, and new key logins.
|
|
169
|
+
* The value is the operator-facing reason; it is registry-only and never seeded into saved
|
|
170
|
+
* config.
|
|
171
|
+
*/
|
|
172
|
+
deprecated?: string;
|
|
163
173
|
note?: string;
|
|
164
174
|
dashboardUrl?: string;
|
|
165
175
|
defaultModel?: string;
|
|
@@ -308,6 +318,15 @@ export interface ProviderRegistryEntry {
|
|
|
308
318
|
*/
|
|
309
319
|
modelDisplayNames?: Record<string, string>;
|
|
310
320
|
modelInputModalities?: Record<string, string[]>;
|
|
321
|
+
/**
|
|
322
|
+
* Provider-wide input modalities for every model `modelInputModalities` does not enumerate,
|
|
323
|
+
* including ids that arrive from live discovery. Declare it where the limit belongs to the
|
|
324
|
+
* provider's API rather than to one model; a per-model entry still wins over it. Registry-only
|
|
325
|
+
* and runtime-only: never seeded or enriched into saved config, and applied only while a provider
|
|
326
|
+
* points at this entry's destination (routed provider, vision eligibility).
|
|
327
|
+
*/
|
|
328
|
+
inputModalities?: string[];
|
|
329
|
+
modelCapabilities?: OcxProviderConfig["modelCapabilities"];
|
|
311
330
|
defaultMaxOutputTokens?: number;
|
|
312
331
|
modelMaxOutputTokens?: Record<string, number>;
|
|
313
332
|
reasoningEfforts?: string[];
|
|
@@ -334,6 +353,8 @@ export interface ProviderRegistryEntry {
|
|
|
334
353
|
* `directReasoningEffortModels` above is registry-owned.
|
|
335
354
|
*/
|
|
336
355
|
noJsonSchemaModels?: string[];
|
|
356
|
+
/** Exact models with no established structured-output wire format. Registry-owned at routing. */
|
|
357
|
+
noStructuredOutputModels?: string[];
|
|
337
358
|
/** Opt this provider into parallel tool calls (see OcxProviderConfig.parallelToolCalls). */
|
|
338
359
|
parallelToolCalls?: boolean;
|
|
339
360
|
/** Opt this provider into forwarding prompt_cache_key (OpenAI-specific; strict backends reject it). */
|
|
@@ -346,6 +367,12 @@ export interface ProviderRegistryEntry {
|
|
|
346
367
|
chatServiceTier?: boolean;
|
|
347
368
|
/** OpenAI Chat EOF policy for gateways that omit terminal frames after complete tool calls. */
|
|
348
369
|
openaiChatEofTolerance?: boolean;
|
|
370
|
+
/**
|
|
371
|
+
* Record that this destination rejects the Chat `developer` role (`400 role 'developer' is not
|
|
372
|
+
* allowed`), so both the translated route and the native `/v1/chat/completions` passthrough send
|
|
373
|
+
* `system` instead. Seeds `foldDeveloperRoleToSystem` on the provider config; a user value wins.
|
|
374
|
+
*/
|
|
375
|
+
foldDeveloperRoleToSystem?: boolean;
|
|
349
376
|
autoToolChoiceOnlyModels?: string[];
|
|
350
377
|
preserveReasoningContentModels?: string[];
|
|
351
378
|
requiresReasoningPlaceholderModels?: string[];
|
|
@@ -378,10 +405,10 @@ export type ProviderConfigSeed = Pick<
|
|
|
378
405
|
"anthropicOAuthInstance" |
|
|
379
406
|
"adapter" | "baseUrl" | "apiKeyTransport" | "responsesPath" | "chatCompletionsPath" | "authMode" | "keyOptional" | "freeTier" | "modelSuffixBracketStrip" | "defaultModel" | "models"
|
|
380
407
|
| "liveModels" | "contextWindow" | "modelContextWindows" | "modelInputModalities"
|
|
381
|
-
| "modelDisplayNames"
|
|
408
|
+
| "modelDisplayNames" | "modelCapabilities"
|
|
382
409
|
| "modelMaxInputTokens" | "defaultMaxOutputTokens" | "modelMaxOutputTokens"
|
|
383
410
|
| "reasoningEfforts" | "modelReasoningEfforts" | "modelDefaultReasoningEfforts" | "reasoningEffortMap" | "modelReasoningEffortMap" | "reasoningWireFormat"
|
|
384
411
|
| "noVisionModels" | "noReasoningModels" | "noTemperatureModels" | "noTopPModels" | "noStopModels" | "noPenaltyModels"
|
|
385
|
-
| "autoToolChoiceOnlyModels" | "preserveReasoningContentModels" | "requiresReasoningPlaceholderModels" | "reasoningSplitModels" | "inlineThinkTagModels" | "reasoningDetailsModels" | "thinkingToggleModels" | "thinkingBudgetModels" | "escapeBuiltinToolNames" | "openaiChatEofTolerance" | "showThinkingSummary"
|
|
412
|
+
| "autoToolChoiceOnlyModels" | "preserveReasoningContentModels" | "requiresReasoningPlaceholderModels" | "reasoningSplitModels" | "inlineThinkTagModels" | "reasoningDetailsModels" | "thinkingToggleModels" | "thinkingBudgetModels" | "escapeBuiltinToolNames" | "openaiChatEofTolerance" | "foldDeveloperRoleToSystem" | "showThinkingSummary"
|
|
386
413
|
| "googleMode" | "project" | "location" | "headers"
|
|
387
414
|
>;
|
|
@@ -1,5 +1,6 @@
|
|
|
1
|
-
import type { ModelCapabilities, OcxProviderConfig } from "../types";
|
|
2
|
-
import {
|
|
1
|
+
import type { ModelCapabilities, OcxProviderConfig, ReasoningCapabilities } from "../types";
|
|
2
|
+
import { declaredModelCapabilities, mergeReasoningModelCapabilities, reasoningCapabilitiesFromConfig, REASONING_CAPABILITY_FIELDS } from "./reasoning-capabilities";
|
|
3
|
+
import { MODEL_ADAPTER_OVERRIDE_ALLOWED, modelInList, pinnedWireAdapter } from "../types";
|
|
3
4
|
import { isCanonicalOpenAiForwardProvider } from "./openai-tiers";
|
|
4
5
|
import { resolveProviderAuthTransport } from "./fastwire";
|
|
5
6
|
import { fastSwitchOff } from "./fast-opt-in";
|
|
@@ -31,8 +32,8 @@ export type StaticProviderPolicyField =
|
|
|
31
32
|
| "modelMaxOutputTokens" | "reasoningEfforts" | "modelReasoningEfforts" | "modelReasoningEffortsAuthoritative"
|
|
32
33
|
| "modelDefaultReasoningEfforts" | "reasoningEffortMap" | "modelReasoningEffortMap"
|
|
33
34
|
| "reasoningWireFormat" | "noVisionModels" | "noReasoningModels" | "noTemperatureModels"
|
|
34
|
-
| "noTopPModels" | "noStopModels" | "noPenaltyModels" | "noJsonSchemaModels" | "parallelToolCalls"
|
|
35
|
-
| "promptCacheKey" | "chatServiceTier" | "openaiChatEofTolerance" | "statelessResponses"
|
|
35
|
+
| "noTopPModels" | "noStopModels" | "noPenaltyModels" | "noJsonSchemaModels" | "noStructuredOutputModels" | "parallelToolCalls"
|
|
36
|
+
| "promptCacheKey" | "chatServiceTier" | "openaiChatEofTolerance" | "foldDeveloperRoleToSystem" | "statelessResponses"
|
|
36
37
|
| "requiresAdjacentResponsesToolResults" | "requiresPairedResponsesToolResults" | "annotateEmptyToolOutputs"
|
|
37
38
|
| "fastWire" | "supportsServiceTier" | "modelSupportsServiceTier" | "supportsOpenAiWebSearchToolFields"
|
|
38
39
|
| "supportsResponsesCustomTools" | "preserveResponsesReasoningContent" | "dropResponsesReasoningItems"
|
|
@@ -48,7 +49,7 @@ export type ResolvedProviderStaticPolicy = Readonly<Partial<StaticProviderPolicy
|
|
|
48
49
|
adapter: string;
|
|
49
50
|
baseUrl: string;
|
|
50
51
|
}>;
|
|
51
|
-
export interface ResolvedPerModelStaticPolicy {
|
|
52
|
+
export interface ResolvedPerModelStaticPolicy extends Readonly<ReasoningCapabilities> {
|
|
52
53
|
readonly adapter: string;
|
|
53
54
|
readonly contextWindow?: number;
|
|
54
55
|
readonly inputModalities?: readonly string[];
|
|
@@ -173,7 +174,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
173
174
|
"responsesPath", "chatCompletionsPath", "keyOptional", "freeTier", "modelSuffixBracketStrip",
|
|
174
175
|
"defaultModel", "models", "liveModels", "contextWindow", "defaultMaxOutputTokens",
|
|
175
176
|
"reasoningWireFormat", "parallelToolCalls", "promptCacheKey",
|
|
176
|
-
"openaiChatEofTolerance", "statelessResponses",
|
|
177
|
+
"openaiChatEofTolerance", "foldDeveloperRoleToSystem", "statelessResponses",
|
|
177
178
|
"requiresAdjacentResponsesToolResults", "requiresPairedResponsesToolResults",
|
|
178
179
|
"annotateEmptyToolOutputs", "fastWire", "supportsServiceTier",
|
|
179
180
|
"supportsOpenAiWebSearchToolFields", "supportsResponsesCustomTools",
|
|
@@ -215,9 +216,11 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
215
216
|
putMergedMap("modelSupportsReasoningSummaries", entry?.modelSupportsReasoningSummaries, provider.modelSupportsReasoningSummaries);
|
|
216
217
|
putMergedMap("modelSupportsVerbosity", entry?.modelSupportsVerbosity, provider.modelSupportsVerbosity);
|
|
217
218
|
for (const key of [
|
|
218
|
-
"
|
|
219
|
+
"modelAutoCompactTokenLimits", "modelSuppressSyntheticMax",
|
|
219
220
|
"modelReasoningSummaryDelivery", "codexToolMode", "modelAdapters",
|
|
220
221
|
] as const) putScalar(key, undefined);
|
|
222
|
+
put("modelCapabilities", mergeReasoningModelCapabilities(entry?.modelCapabilities, provider.modelCapabilities),
|
|
223
|
+
provider.modelCapabilities !== undefined ? "operator-capability" : entry?.modelCapabilities !== undefined ? "registry" : "unknown");
|
|
221
224
|
for (const key of ["modelContextWindows", "modelMaxInputTokens"] as const) {
|
|
222
225
|
const [value, source] = input.providerName === "openai-apikey"
|
|
223
226
|
? positiveCapMap(entry?.[key], provider[key])
|
|
@@ -231,7 +234,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
231
234
|
for (const key of [
|
|
232
235
|
"noVisionModels", "noReasoningModels", "noTemperatureModels", "noTopPModels",
|
|
233
236
|
"noStopModels",
|
|
234
|
-
"noPenaltyModels", "noJsonSchemaModels", "autoToolChoiceOnlyModels",
|
|
237
|
+
"noPenaltyModels", "noJsonSchemaModels", "noStructuredOutputModels", "autoToolChoiceOnlyModels",
|
|
235
238
|
"preserveReasoningContentModels", "requiresReasoningPlaceholderModels",
|
|
236
239
|
"reasoningSplitModels", "reasoningDetailsModels", "thinkingToggleModels", "thinkingBudgetModels",
|
|
237
240
|
] as const) putUnion(key, entry?.[key]);
|
|
@@ -325,7 +328,16 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
325
328
|
: exactMaxInputTokens;
|
|
326
329
|
const exactMaxOutputTokens = modelValue(providerPolicy.modelMaxOutputTokens);
|
|
327
330
|
const modelMaxOutputTokens = exactMaxOutputTokens ?? providerPolicy.defaultMaxOutputTokens;
|
|
328
|
-
const
|
|
331
|
+
const reasoningCapabilities = reasoningCapabilitiesFromConfig({
|
|
332
|
+
...providerPolicy,
|
|
333
|
+
adapter: providerPolicy.adapter ?? provider.adapter,
|
|
334
|
+
baseUrl: providerPolicy.baseUrl ?? provider.baseUrl,
|
|
335
|
+
...(input.modelCapabilities ? { modelCapabilities: mergeReasoningModelCapabilities(
|
|
336
|
+
providerPolicy.modelCapabilities, { [input.modelId]: input.modelCapabilities },
|
|
337
|
+
) } : {}),
|
|
338
|
+
}, input.modelId);
|
|
339
|
+
const modelReasoningEfforts = reasoningCapabilities.supportsReasoningEffort === false ? []
|
|
340
|
+
: modelValue(providerPolicy.modelReasoningEfforts) ?? providerPolicy.reasoningEfforts;
|
|
329
341
|
const modelSupportsReasoningSummaries = modelValue(providerPolicy.modelSupportsReasoningSummaries)
|
|
330
342
|
?? (modelValue(providerPolicy.modelReasoningSummaryDelivery) !== undefined ? true : undefined);
|
|
331
343
|
const modelSupportsVerbosity = modelValue(providerPolicy.modelSupportsVerbosity) ?? providerPolicy.supportsVerbosity;
|
|
@@ -334,6 +346,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
334
346
|
: modelValue(providerPolicy.modelSupportsServiceTier) ?? providerPolicy.supportsServiceTier;
|
|
335
347
|
const model: ResolvedPerModelStaticPolicy = {
|
|
336
348
|
adapter,
|
|
349
|
+
...reasoningCapabilities,
|
|
337
350
|
...(modelContextWindow !== undefined ? { contextWindow: modelContextWindow } : {}),
|
|
338
351
|
...(capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) !== undefined
|
|
339
352
|
? { inputModalities: capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) }
|
|
@@ -341,7 +354,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
341
354
|
...(modelMaxInputTokens !== undefined ? { maxInputTokens: modelMaxInputTokens } : {}),
|
|
342
355
|
...(modelMaxOutputTokens !== undefined ? { maxOutputTokens: modelMaxOutputTokens } : {}),
|
|
343
356
|
...(modelReasoningEfforts !== undefined ? { reasoningEfforts: modelReasoningEfforts } : {}),
|
|
344
|
-
...(modelValue(providerPolicy.modelDefaultReasoningEfforts) !== undefined ? { defaultReasoningEffort: modelValue(providerPolicy.modelDefaultReasoningEfforts) } : {}),
|
|
357
|
+
...(reasoningCapabilities.supportsReasoningEffort !== false && modelValue(providerPolicy.modelDefaultReasoningEfforts) !== undefined ? { defaultReasoningEffort: modelValue(providerPolicy.modelDefaultReasoningEfforts) } : {}),
|
|
345
358
|
...(modelSupportsReasoningSummaries !== undefined ? { supportsReasoningSummaries: modelSupportsReasoningSummaries } : {}),
|
|
346
359
|
...(modelSupportsVerbosity !== undefined ? { supportsVerbosity: modelSupportsVerbosity } : {}),
|
|
347
360
|
...(modelSupportsServiceTier !== undefined ? { supportsServiceTier: modelSupportsServiceTier } : {}),
|
|
@@ -359,6 +372,13 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
|
|
|
359
372
|
const modelProvenance: Partial<Record<keyof ResolvedPerModelStaticPolicy, StaticPolicySource>> = {
|
|
360
373
|
adapter: adapterSource,
|
|
361
374
|
};
|
|
375
|
+
for (const field of REASONING_CAPABILITY_FIELDS) {
|
|
376
|
+
modelProvenance[field] = reasoningCapabilities[field] === undefined ? "unknown"
|
|
377
|
+
: modelInList(provider.noReasoningModels, input.modelId) ? "operator"
|
|
378
|
+
: modelInList(entry?.noReasoningModels, input.modelId) ? "registry"
|
|
379
|
+
: input.modelCapabilities?.[field] !== undefined || declaredModelCapabilities(provider.modelCapabilities, input.modelId)?.[field] !== undefined
|
|
380
|
+
? "operator-capability" : declaredModelCapabilities(entry?.modelCapabilities, input.modelId)?.[field] !== undefined ? "registry" : "operator";
|
|
381
|
+
}
|
|
362
382
|
modelProvenance.contextWindow = modelOrProviderSource(provider.modelContextWindows, entry?.modelContextWindows, provider.contextWindow, entry?.contextWindow);
|
|
363
383
|
if (modelProvenance.contextWindow === "unknown" && familyContextWindow !== undefined) {
|
|
364
384
|
modelProvenance.contextWindow = anthropicFamilyContextWindow(provider.modelContextWindows, input.modelId) !== undefined
|
|
@@ -13,7 +13,7 @@ import { captureProviderApiKeySelection } from "./api-key-selection-capture";
|
|
|
13
13
|
import { assertProviderDestinationAllowed } from "../lib/destination-policy";
|
|
14
14
|
import { redactSecretString, redactUrlForLog } from "../lib/redact";
|
|
15
15
|
import { PROVIDER_REGISTRY, registryEntryForProviderDestination } from "./registry";
|
|
16
|
-
import { applyDirectReasoningEffortContracts, hasLegacyClinePassReasoningEfforts } from "./derive";
|
|
16
|
+
import { applyDirectReasoningEffortContracts, hasLegacyClinePassReasoningEfforts, registryInputModalitiesForDestination } from "./derive";
|
|
17
17
|
import { cloneFastWire } from "./fastwire";
|
|
18
18
|
import { fastSwitchOff } from "./fast-opt-in";
|
|
19
19
|
import {
|
|
@@ -21,6 +21,7 @@ import {
|
|
|
21
21
|
providerSupportsLiveModelDiscovery,
|
|
22
22
|
} from "./static-model-discovery";
|
|
23
23
|
import { resolveModelPolicy } from "./resolved-model-policy";
|
|
24
|
+
import { mapFill, nestedMapFill, stableUnion } from "./resolved-model-policy-merge";
|
|
24
25
|
|
|
25
26
|
/** Same endpoint modulo surrounding space and trailing slashes — matches `matchBaseUrlChoice`. */
|
|
26
27
|
function isSameEndpoint(a: string, b: string): boolean {
|
|
@@ -121,9 +122,21 @@ export function routedProviderConfig(providerName: string, provider: OcxProvider
|
|
|
121
122
|
// the reasoning item the upstream answers 400 to. Destination matching refuses templated and
|
|
122
123
|
// overridable base URLs, so this cannot follow a retargeted row, and an explicit value wins.
|
|
123
124
|
const destination = registryEntryForProviderDestination(provider);
|
|
125
|
+
const inputModalities = provider.inputModalities === undefined
|
|
126
|
+
? registryInputModalitiesForDestination(providerName, provider) : undefined;
|
|
124
127
|
return {
|
|
125
128
|
...provider,
|
|
129
|
+
...(inputModalities ? { inputModalities } : {}),
|
|
126
130
|
apiKey: usableResolvedApiKey(provider.apiKey),
|
|
131
|
+
...(destination?.modelDefaultReasoningEfforts ? {
|
|
132
|
+
modelDefaultReasoningEfforts: mapFill(destination.modelDefaultReasoningEfforts, provider.modelDefaultReasoningEfforts)[0],
|
|
133
|
+
} : {}),
|
|
134
|
+
...(destination?.modelReasoningEffortMap ? {
|
|
135
|
+
modelReasoningEffortMap: nestedMapFill(destination.modelReasoningEffortMap, provider.modelReasoningEffortMap)[0],
|
|
136
|
+
} : {}),
|
|
137
|
+
...(destination?.noStructuredOutputModels?.length ? {
|
|
138
|
+
noStructuredOutputModels: stableUnion(destination.noStructuredOutputModels, provider.noStructuredOutputModels)[0],
|
|
139
|
+
} : {}),
|
|
127
140
|
...(provider.dropResponsesReasoningItems === undefined && destination?.dropResponsesReasoningItems !== undefined
|
|
128
141
|
? { dropResponsesReasoningItems: destination.dropResponsesReasoningItems }
|
|
129
142
|
: {}),
|
|
@@ -172,6 +185,7 @@ export function routedProviderConfig(providerName: string, provider: OcxProvider
|
|
|
172
185
|
const noStopModels = staticPolicy.noStopModels;
|
|
173
186
|
const noPenaltyModels = staticPolicy.noPenaltyModels;
|
|
174
187
|
const noJsonSchemaModels = staticPolicy.noJsonSchemaModels;
|
|
188
|
+
const noStructuredOutputModels = staticPolicy.noStructuredOutputModels;
|
|
175
189
|
const autoToolChoiceOnlyModels = staticPolicy.autoToolChoiceOnlyModels;
|
|
176
190
|
const preserveReasoningContentModels = staticPolicy.preserveReasoningContentModels;
|
|
177
191
|
const requiresReasoningPlaceholderModels = staticPolicy.requiresReasoningPlaceholderModels;
|
|
@@ -287,6 +301,11 @@ export function routedProviderConfig(providerName: string, provider: OcxProvider
|
|
|
287
301
|
...(provider.project === undefined && registryEntry.project !== undefined ? { project: registryEntry.project } : {}),
|
|
288
302
|
...(provider.location === undefined && registryEntry.location !== undefined ? { location: registryEntry.location } : {}),
|
|
289
303
|
...(provider.contextWindow === undefined && registryEntry.contextWindow !== undefined ? { contextWindow: registryEntry.contextWindow } : {}),
|
|
304
|
+
// The request path never calls enrichProviderFromRegistry(), so a row saved before the
|
|
305
|
+
// registry declared a provider-wide modality must inherit it here to reach the vision gate.
|
|
306
|
+
...(provider.inputModalities === undefined && registryEntry.inputModalities !== undefined
|
|
307
|
+
? { inputModalities: [...registryEntry.inputModalities] }
|
|
308
|
+
: {}),
|
|
290
309
|
...((provider.reasoningEfforts === undefined || hasLegacyClinePassReasoningEfforts(providerName, provider))
|
|
291
310
|
&& registryEntry.reasoningEfforts !== undefined
|
|
292
311
|
? { reasoningEfforts: [...registryEntry.reasoningEfforts] }
|
|
@@ -302,6 +321,9 @@ export function routedProviderConfig(providerName: string, provider: OcxProvider
|
|
|
302
321
|
...(provider.openaiChatEofTolerance === undefined && registryEntry.openaiChatEofTolerance !== undefined
|
|
303
322
|
? { openaiChatEofTolerance: registryEntry.openaiChatEofTolerance }
|
|
304
323
|
: {}),
|
|
324
|
+
...(provider.foldDeveloperRoleToSystem === undefined && registryEntry.foldDeveloperRoleToSystem !== undefined
|
|
325
|
+
? { foldDeveloperRoleToSystem: registryEntry.foldDeveloperRoleToSystem }
|
|
326
|
+
: {}),
|
|
305
327
|
...(provider.reasoningWireFormat === undefined && registryEntry.reasoningWireFormat !== undefined
|
|
306
328
|
? { reasoningWireFormat: registryEntry.reasoningWireFormat }
|
|
307
329
|
: {}),
|
|
@@ -325,6 +347,7 @@ export function routedProviderConfig(providerName: string, provider: OcxProvider
|
|
|
325
347
|
...(noStopModels ? { noStopModels } : {}),
|
|
326
348
|
...(noPenaltyModels ? { noPenaltyModels } : {}),
|
|
327
349
|
...(noJsonSchemaModels ? { noJsonSchemaModels } : {}),
|
|
350
|
+
...(noStructuredOutputModels ? { noStructuredOutputModels } : {}),
|
|
328
351
|
...(autoToolChoiceOnlyModels ? { autoToolChoiceOnlyModels } : {}),
|
|
329
352
|
...(preserveReasoningContentModels ? { preserveReasoningContentModels } : {}),
|
|
330
353
|
...(requiresReasoningPlaceholderModels ? { requiresReasoningPlaceholderModels } : {}),
|