@bitkyc08/opencodex 2.92.0 → 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.
Files changed (259) hide show
  1. package/bin/ocx.mjs +10 -5
  2. package/gui/dist/assets/App-CJZoq-uF.js +58 -0
  3. package/gui/dist/assets/{Tray-BSsxznDm.js → Tray-BWQPcgZn.js} +1 -1
  4. package/gui/dist/assets/index-DRe8Hhfz.css +1 -0
  5. package/gui/dist/assets/{index-BlFE_VB_.js → index-Xfct8owH.js} +12 -12
  6. package/gui/dist/assets/{quota-summary-BRlbXgw0.js → quota-summary-DxToX1rV.js} +1 -1
  7. package/gui/dist/index.html +2 -2
  8. package/package.json +5 -5
  9. package/src/adapters/anthropic.ts +28 -5
  10. package/src/adapters/coding-agent/batch-dispatcher.ts +122 -0
  11. package/src/adapters/coding-agent/profile.ts +54 -16
  12. package/src/adapters/coding-agent/turn.ts +30 -2
  13. package/src/adapters/command-code.ts +114 -7
  14. package/src/adapters/cursor/native-exec-fs.ts +11 -9
  15. package/src/adapters/devin/live-models.ts +4 -0
  16. package/src/adapters/google-wire-compiler.ts +1 -2
  17. package/src/adapters/kiro/reasoning.ts +2 -0
  18. package/src/adapters/kiro-refusal.ts +6 -1
  19. package/src/adapters/kiro-wire.ts +1 -1
  20. package/src/adapters/openai-chat/messages.ts +8 -2
  21. package/src/adapters/openai-chat/passthrough.ts +35 -14
  22. package/src/adapters/openai-chat/reasoning-replay.ts +27 -16
  23. package/src/adapters/openai-chat/reasoning-wire.ts +117 -8
  24. package/src/adapters/openai-chat/required-tool-result.ts +64 -0
  25. package/src/adapters/openai-chat.ts +7 -8
  26. package/src/adapters/openai-responses/passthrough.ts +7 -4
  27. package/src/adapters/openai-responses/reasoning.ts +13 -0
  28. package/src/adapters/openai-responses/request-strips.ts +1 -1
  29. package/src/adapters/qoder/adapter.ts +129 -8
  30. package/src/adapters/qoder/scaffold-guard.ts +5 -0
  31. package/src/adapters/reasoning-controls.ts +16 -0
  32. package/src/adapters/tool-call-id.ts +1 -3
  33. package/src/bridge/errors.ts +27 -0
  34. package/src/bridge.ts +1 -1
  35. package/src/chat/inbound.ts +4 -1
  36. package/src/chat/outbound.ts +9 -2
  37. package/src/chat/reasoning-details.ts +28 -5
  38. package/src/claude/desktop-first-party.ts +54 -4
  39. package/src/claude/desktop-picker.ts +11 -8
  40. package/src/claude/first-party-settings.ts +3 -1
  41. package/src/claude/inbound.ts +2 -1
  42. package/src/claude/intercept/local-ca.ts +2 -1
  43. package/src/claude/intercept/picker-ca-cleanup.ts +4 -2
  44. package/src/claude/intercept/picker-ca-startup.ts +3 -1
  45. package/src/claude/intercept/picker-models.ts +6 -3
  46. package/src/claude/intercept/picker-runtime.ts +3 -2
  47. package/src/claude/intercept/runtime.ts +6 -3
  48. package/src/claude/model-info.ts +31 -8
  49. package/src/cli/account-api.ts +11 -0
  50. package/src/cli/capabilities-agents-routing.ts +2 -2
  51. package/src/cli/capabilities-base.ts +7 -1
  52. package/src/cli/capabilities-observe-system.ts +10 -0
  53. package/src/cli/claude-desktop.ts +9 -8
  54. package/src/cli/claude.ts +3 -2
  55. package/src/cli/codex-shim-autorestore.ts +9 -0
  56. package/src/cli/combo-result.ts +7 -0
  57. package/src/cli/combo.ts +18 -0
  58. package/src/cli/cross-home-owner.ts +13 -1
  59. package/src/cli/dispatch.ts +25 -6
  60. package/src/cli/ensure-desired-integrations.ts +8 -6
  61. package/src/cli/export-command.ts +4 -1
  62. package/src/cli/help.ts +2 -1
  63. package/src/cli/index.ts +45 -45
  64. package/src/cli/integration-plan-dto.ts +4 -0
  65. package/src/cli/opencode.ts +20 -9
  66. package/src/cli/registry.ts +6 -4
  67. package/src/cli/root.ts +9 -0
  68. package/src/cli/status.ts +59 -1
  69. package/src/cli/stop-restore.ts +113 -0
  70. package/src/cli/sync-expectation.ts +60 -0
  71. package/src/cli/uninstall-integrations.ts +41 -7
  72. package/src/client/connect.ts +40 -3
  73. package/src/client/link-home.ts +10 -1
  74. package/src/client/machine-api.ts +3 -1
  75. package/src/client/sync-expectation.ts +118 -0
  76. package/src/clients/config-export/agent-clients.ts +96 -0
  77. package/src/clients/config-export/contracts.ts +25 -7
  78. package/src/clients/config-export/qoder.ts +38 -23
  79. package/src/clients/config-export/reasoning-metadata.ts +55 -8
  80. package/src/clients/config-export.ts +12 -0
  81. package/src/clients/lazycodex.ts +7 -2
  82. package/src/clients/reasoning-controls.ts +49 -0
  83. package/src/codex/auth-api/account-identity.ts +14 -0
  84. package/src/codex/auth-api/account-list.ts +29 -2
  85. package/src/codex/auth-api/http.ts +13 -0
  86. package/src/codex/auth-api/main-account-probe.ts +2 -2
  87. package/src/codex/auth-api/pool-mode-gate.ts +16 -3
  88. package/src/codex/auth-api/pool-quota-probe.ts +15 -0
  89. package/src/codex/auth-context.ts +15 -12
  90. package/src/codex/catalog/build-entries.ts +22 -0
  91. package/src/codex/catalog/configured-tiers.ts +70 -0
  92. package/src/codex/catalog/effort.ts +44 -10
  93. package/src/codex/catalog/gather-capture.ts +1 -0
  94. package/src/codex/catalog/heal-observation.ts +155 -13
  95. package/src/codex/catalog/metadata.ts +6 -0
  96. package/src/codex/catalog/model-hints.ts +60 -11
  97. package/src/codex/catalog/model-visibility.ts +3 -2
  98. package/src/codex/catalog/native-models.ts +18 -3
  99. package/src/codex/catalog/parsing.ts +10 -8
  100. package/src/codex/catalog/pinned-models.ts +2 -1
  101. package/src/codex/catalog/provider-models.ts +4 -5
  102. package/src/codex/catalog/retained-sync.ts +2 -1
  103. package/src/codex/catalog/routed-gather.ts +19 -20
  104. package/src/codex/catalog/write-audit.ts +3 -2
  105. package/src/codex/catalog-self-heal.ts +43 -23
  106. package/src/codex/collision-identity.ts +30 -5
  107. package/src/codex/convergence.ts +4 -3
  108. package/src/codex/data/roster-pinned-models.json +177 -1
  109. package/src/codex/desired-state.ts +6 -1
  110. package/src/codex/desktop-compatibility/installed-build.ts +88 -12
  111. package/src/codex/desktop-compatibility/runtime.ts +29 -5
  112. package/src/codex/desktop-compatibility/usage-controller.ts +2 -0
  113. package/src/codex/desktop-compatibility/windows-certificate-trust.ts +46 -12
  114. package/src/codex/inject/config-toml.ts +0 -18
  115. package/src/codex/inject/plan.ts +18 -3
  116. package/src/codex/inject/remove.ts +6 -2
  117. package/src/codex/inject.ts +1 -1
  118. package/src/codex/journal.ts +61 -1
  119. package/src/codex/model-entitlements.ts +1 -1
  120. package/src/codex/native-main-owner.ts +27 -1
  121. package/src/codex/native-profile-store.ts +2 -1
  122. package/src/codex/pool-credit-policy.ts +35 -1
  123. package/src/codex/quota-types.ts +3 -0
  124. package/src/codex/quota.ts +24 -4
  125. package/src/codex/role-auto-assign.ts +6 -2
  126. package/src/codex/subagent-defaults.ts +49 -4
  127. package/src/combos/failover.ts +17 -2
  128. package/src/combos/jev-decision-contract.ts +22 -1
  129. package/src/combos/jev-service-exchange.ts +3 -1
  130. package/src/combos/jev.ts +95 -9
  131. package/src/combos/resolve.ts +45 -1
  132. package/src/combos/types.ts +24 -0
  133. package/src/config/load-degrade.ts +5 -0
  134. package/src/config/paths.ts +19 -3
  135. package/src/config/pending-teardown.ts +22 -1
  136. package/src/config/provider-validation.ts +15 -4
  137. package/src/config/removed-opencode-free.ts +11 -0
  138. package/src/config/schema/leaf-validators.ts +5 -0
  139. package/src/generated/compatibility-version.json +314 -230
  140. package/src/generated/model-metadata.ts +5 -5
  141. package/src/images/loop.ts +16 -2
  142. package/src/integrations/catalog-refresh.ts +1 -1
  143. package/src/integrations/hermes-profiles.ts +251 -0
  144. package/src/integrations/merge.ts +8 -0
  145. package/src/integrations/mutation-plan.ts +19 -4
  146. package/src/integrations/registry.ts +5 -0
  147. package/src/integrations/state.ts +36 -1
  148. package/src/integrations/writer.ts +18 -4
  149. package/src/lib/bun-path-runtime.mjs +2 -1
  150. package/src/lib/errors.ts +62 -0
  151. package/src/lib/eventstream-decoder.ts +81 -37
  152. package/src/lib/opaque-proxy-tunnel.ts +32 -8
  153. package/src/lib/pinned-http.ts +35 -2
  154. package/src/lib/process-control.ts +13 -1
  155. package/src/lib/socks5-fetch.ts +59 -64
  156. package/src/lib/socks5-upload-waiter.ts +50 -0
  157. package/src/lib/sse-decoder.ts +7 -2
  158. package/src/lib/state-store-sweeper.ts +40 -14
  159. package/src/lib/windows-native-principal-worker.ts +14 -0
  160. package/src/lib/windows-native-principal.ts +188 -0
  161. package/src/lib/windows-secret-acl.ts +39 -3
  162. package/src/lib/windows-user-principal.ts +62 -18
  163. package/src/link/ssh-config.ts +35 -3
  164. package/src/providers/account-quota-disk.ts +12 -4
  165. package/src/providers/codebuddy-models.ts +6 -0
  166. package/src/providers/codex-capacity.ts +3 -1
  167. package/src/providers/derive.ts +28 -0
  168. package/src/providers/kiro-models.ts +3 -0
  169. package/src/providers/mimo-thinking.ts +12 -0
  170. package/src/providers/model-rename-fields.ts +1 -0
  171. package/src/providers/model-rename-migration.ts +10 -0
  172. package/src/providers/openai-sidecar-credit.ts +2 -2
  173. package/src/providers/reasoning-capabilities.ts +76 -0
  174. package/src/providers/registry/entries-core.ts +14 -12
  175. package/src/providers/registry/entries-extended.ts +63 -20
  176. package/src/providers/registry/model-ids.ts +2 -0
  177. package/src/providers/registry/model-seeds.ts +93 -28
  178. package/src/providers/registry/types.ts +10 -1
  179. package/src/providers/resolved-model-policy.ts +26 -6
  180. package/src/providers/routed-config.ts +9 -1
  181. package/src/providers/stale-model-roster-migration.ts +44 -7
  182. package/src/remote-control/workspace-pi-runtime.ts +11 -6
  183. package/src/responses/compaction-summary.ts +45 -0
  184. package/src/responses/muse-tool-name-alias.ts +1 -2
  185. package/src/responses/parser.ts +3 -2
  186. package/src/responses/schema.ts +1 -0
  187. package/src/responses/thought-signature-replay.ts +10 -2
  188. package/src/routing/capability.ts +8 -2
  189. package/src/routing/history/ingest.ts +28 -3
  190. package/src/server/auth-cors.ts +3 -2
  191. package/src/server/chat-completions.ts +7 -0
  192. package/src/server/chat-native-eligibility.ts +7 -0
  193. package/src/server/chat-native-reasoning.ts +14 -4
  194. package/src/server/direct-local-http.ts +5 -2
  195. package/src/server/effort-policy.ts +5 -1
  196. package/src/server/index/live-sideband.ts +18 -15
  197. package/src/server/index/serve-options.ts +5 -4
  198. package/src/server/management/agent-settings-routes.ts +27 -15
  199. package/src/server/management/combo-routes.ts +4 -0
  200. package/src/server/management/config-routes.ts +2 -2
  201. package/src/server/management/decision-discovery.ts +12 -2
  202. package/src/server/management/hermes-profile-routes.ts +156 -0
  203. package/src/server/management/integration-routes.ts +85 -24
  204. package/src/server/management/model-routes.ts +2 -2
  205. package/src/server/management/model-row-export-metadata.ts +25 -6
  206. package/src/server/management/model-rows.ts +15 -20
  207. package/src/server/management/native-integration-routes.ts +11 -8
  208. package/src/server/management/route-registry.ts +4 -0
  209. package/src/server/management-api.ts +5 -0
  210. package/src/server/models-capabilities.ts +21 -3
  211. package/src/server/responses/adapter-delivery.ts +11 -1
  212. package/src/server/responses/adapter-dispatch.ts +2 -2
  213. package/src/server/responses/codex-ws-exchange.ts +39 -9
  214. package/src/server/responses/codex-ws-session.ts +38 -4
  215. package/src/server/responses/codex-ws-wire.ts +6 -1
  216. package/src/server/responses/combo-required-tool-contract.ts +103 -0
  217. package/src/server/responses/combo-stream-preflight.ts +8 -1
  218. package/src/server/responses/compact.ts +4 -3
  219. package/src/server/responses/context-overflow.ts +8 -1
  220. package/src/server/responses/core-combo-failure.ts +10 -0
  221. package/src/server/responses/core-combo.ts +23 -15
  222. package/src/server/responses/core-lifetime.ts +3 -0
  223. package/src/server/responses/core-normalize.ts +11 -1
  224. package/src/server/responses/core-options.ts +2 -0
  225. package/src/server/responses/main-hard-lock-routing.ts +60 -44
  226. package/src/server/responses/passthrough-delivery.ts +40 -2
  227. package/src/server/responses/policy-fallback.ts +6 -1
  228. package/src/server/responses/request-compression.ts +12 -7
  229. package/src/server/responses/request-prepare.ts +22 -2
  230. package/src/server/responses/request-sidecar-auth.ts +2 -0
  231. package/src/server/responses/response-effects.ts +22 -10
  232. package/src/server/responses/sidecar-execution.ts +11 -3
  233. package/src/server/stop-teardown.ts +36 -1
  234. package/src/server/ws-bridge.ts +1 -1
  235. package/src/service/desktop-supervision.d.mts +7 -4
  236. package/src/service/desktop-supervision.mjs +90 -13
  237. package/src/service/ownership-mutation-lease.mjs +21 -7
  238. package/src/types/config.ts +8 -2
  239. package/src/types/provider.ts +20 -6
  240. package/src/types/request.ts +10 -0
  241. package/src/types.ts +1 -0
  242. package/src/update/bind-probe.ts +29 -0
  243. package/src/update/job.ts +10 -29
  244. package/src/update/npm-install-job.ps1 +173 -0
  245. package/src/update/npm-install.mjs +70 -0
  246. package/src/update/stop-contract.d.mts +1 -0
  247. package/src/update/stop-contract.mjs +14 -0
  248. package/src/update/stop-decision.d.mts +1 -1
  249. package/src/update/stop-decision.mjs +13 -3
  250. package/src/update/teardown-handover.ts +166 -0
  251. package/src/update/transactional-install.d.mts +14 -3
  252. package/src/update/transactional-install.mjs +33 -7
  253. package/src/usage/expected-prices.ts +64 -31
  254. package/src/usage/timeline.ts +19 -9
  255. package/src/vision/eligibility.ts +15 -5
  256. package/src/vision/plan.ts +2 -2
  257. package/src/web-search/loop.ts +16 -2
  258. package/gui/dist/assets/App-sUf_8afJ.js +0 -58
  259. package/gui/dist/assets/index-DPeB9HpV.css +0 -1
@@ -29,6 +29,8 @@ const FILENAME = "provider-account-quota-cache.json";
29
29
  */
30
30
  const DISK_MAX_AGE_MS = 6 * 60 * 60_000;
31
31
  const PERSIST_DEBOUNCE_MS = 250;
32
+ /** Match the reset ledger's maximum deferral under sustained quota observations. */
33
+ const MAX_PERSIST_DEFERRAL_MS = 1_000;
32
34
 
33
35
  type DiskFile = {
34
36
  version: 1;
@@ -38,6 +40,7 @@ type DiskFile = {
38
40
  };
39
41
 
40
42
  let persistTimer: ReturnType<typeof setTimeout> | null = null;
43
+ let persistStartedAt: number | null = null;
41
44
 
42
45
  /** Read the snapshot. Returns an empty map for a missing, corrupt or stale file. */
43
46
  export function readPersistedAccountQuotas(now = Date.now()): Map<string, ProviderQuota> {
@@ -86,14 +89,19 @@ export function readPersistedKiroVerdicts(now = Date.now()): Map<string, KiroPer
86
89
  return result;
87
90
  }
88
91
 
89
- /** Write the snapshot, debounced. Best-effort: a failed write is not an error. */
92
+ /** Debounce snapshots without starving writes under load. Failures remain best-effort. */
90
93
  export function schedulePersistAccountQuotas(
91
94
  rows: () => Iterable<[string, ProviderQuota | KiroPersistedQuota]>,
92
95
  verdicts: () => Iterable<[string, KiroPersistedVerdict]> = () => [],
93
96
  ): void {
97
+ const now = performance.now();
98
+ persistStartedAt ??= now;
99
+ const delay = Math.max(0, Math.min(PERSIST_DEBOUNCE_MS,
100
+ MAX_PERSIST_DEFERRAL_MS - (now - persistStartedAt)));
94
101
  if (persistTimer) clearTimeout(persistTimer);
95
102
  persistTimer = setTimeout(() => {
96
103
  persistTimer = null;
104
+ persistStartedAt = null;
97
105
  try {
98
106
  const out: Record<string, ProviderQuota> = {};
99
107
  for (const [key, quota] of rows()) out[key] = quota;
@@ -106,12 +114,12 @@ export function schedulePersistAccountQuotas(
106
114
  } catch {
107
115
  // Best-effort persistence only.
108
116
  }
109
- }, PERSIST_DEBOUNCE_MS);
117
+ }, delay);
110
118
  }
111
119
 
112
120
  /** Test seam: drop any pending write so a suite cannot leak one into the next file. */
113
121
  export function cancelPendingAccountQuotaPersist(): void {
114
- if (!persistTimer) return;
115
- clearTimeout(persistTimer);
122
+ if (persistTimer) clearTimeout(persistTimer);
116
123
  persistTimer = null;
124
+ persistStartedAt = null;
117
125
  }
@@ -30,6 +30,8 @@ export const CODEBUDDY_GLOBAL_MODELS = [
30
30
  "gpt-6-luna",
31
31
  // 260930 preemptive: GPT-6.1 Sol (OpenAI announced 2026-09-29); mirrors the GPT-6 Sol row.
32
32
  "gpt-6.1-sol",
33
+ // 261027 preemptive: GPT-6.1 Terra (mirrors CodeBuddy's GPT-6.1 Sol seed; provider listing unverified).
34
+ "gpt-6.1-terra",
33
35
  "gpt-5.5",
34
36
  "gpt-5.4",
35
37
  "gpt-5.3-codex",
@@ -82,6 +84,7 @@ export const CODEBUDDY_GLOBAL_MODEL_CONTEXT_WINDOWS: Record<string, number> = {
82
84
  "gpt-6-sol": 1_000_000,
83
85
  "gpt-6-luna": 1_000_000,
84
86
  "gpt-6.1-sol": 1_000_000,
87
+ "gpt-6.1-terra": 1_000_000,
85
88
  "gpt-5.5": 1_000_000,
86
89
  "gpt-5.4": 272_000,
87
90
  "gpt-5.3-codex": 272_000,
@@ -105,6 +108,7 @@ export const CODEBUDDY_GLOBAL_MODEL_MAX_OUTPUT_TOKENS: Record<string, number> =
105
108
  "gpt-6-sol": 128_000,
106
109
  "gpt-6-luna": 128_000,
107
110
  "gpt-6.1-sol": 128_000,
111
+ "gpt-6.1-terra": 128_000,
108
112
  "gpt-5.5": 72_000,
109
113
  "gpt-5.4": 128_000,
110
114
  "gpt-5.3-codex": 128_000,
@@ -124,6 +128,7 @@ export const CODEBUDDY_GLOBAL_MODEL_REASONING_EFFORTS: Record<string, string[]>
124
128
  "gpt-6-sol": ["low", "medium", "high", "xhigh"],
125
129
  "gpt-6-luna": ["low", "medium", "high", "xhigh"],
126
130
  "gpt-6.1-sol": ["low", "medium", "high", "xhigh"],
131
+ "gpt-6.1-terra": ["low", "medium", "high", "xhigh"],
127
132
  "glm-5.3": ["low", "high", "max"],
128
133
  "glm-5.2": ["high", "xhigh"],
129
134
  };
@@ -135,6 +140,7 @@ export const CODEBUDDY_GLOBAL_MODEL_DEFAULT_REASONING_EFFORTS: Record<string, st
135
140
  "gpt-6-sol": "high",
136
141
  "gpt-6-luna": "high",
137
142
  "gpt-6.1-sol": "high",
143
+ "gpt-6.1-terra": "high",
138
144
  "glm-5.3": "high",
139
145
  "glm-5.2": "high",
140
146
  };
@@ -5,6 +5,8 @@ export const CODEX_CONFIGURED_CAPACITY_WEIGHTS = {
5
5
  team: 1,
6
6
  business: 1,
7
7
  prolite: 5,
8
+ // Business Premium has 5x Standard included usage; this is display calibration only (#7440).
9
+ self_serve_business_prolite: 5,
8
10
  pro: 20,
9
11
  } as const;
10
12
 
@@ -13,7 +15,7 @@ export const CODEX_CONFIGURED_CAPACITY_WEIGHTS = {
13
15
  *
14
16
  * `CodexAccount.plan` is an unrestricted upstream string and cannot be enumerated: the
15
17
  * bundled model snapshot alone carries 21 distinct plan names (`edu_plus`, `finserv`,
16
- * `k12`, `quorum`, `self_serve_business_usage_based`, …) against the five listed above.
18
+ * `k12`, `quorum`, `self_serve_business_usage_based`, …) against the calibrated entries above.
17
19
  * Treating an unlisted plan as unknown dropped the account from the estimate entirely and
18
20
  * reported "incomplete coverage" — which is how a Business seat upgraded to Premium
19
21
  * disappeared from its own capacity report (#3155), and it was already happening to the
@@ -4,8 +4,10 @@ import { cloneFastWire } from "./fastwire";
4
4
  import { isAzureModelMetadataDestination, publishedAzureModelMetadata, type AzureModelMetadata } from "./azure-model-metadata";
5
5
  import { resolveModelPolicy } from "./resolved-model-policy";
6
6
  import { mapFill, nestedMapFill, stableUnion } from "./resolved-model-policy-merge";
7
+ import { mergeReasoningModelCapabilities } from "./reasoning-capabilities";
7
8
  import {
8
9
  PROVIDER_REGISTRY,
10
+ normalizedProviderEndpoint,
9
11
  registryEntryForProviderDestination,
10
12
  registryModelServiceTierCapabilityApplies,
11
13
  type ProviderRegistryEntry,
@@ -31,6 +33,7 @@ export interface DerivedKeyLoginProvider {
31
33
  contextWindow?: number;
32
34
  modelContextWindows?: Record<string, number>;
33
35
  modelInputModalities?: Record<string, string[]>;
36
+ modelCapabilities?: OcxProviderConfig["modelCapabilities"];
34
37
  modelMaxInputTokens?: Record<string, number>;
35
38
  defaultMaxOutputTokens?: number;
36
39
  modelMaxOutputTokens?: Record<string, number>;
@@ -263,6 +266,7 @@ export function providerConfigSeed(entry: ProviderRegistryEntry): OcxProviderCon
263
266
  ...(staticPolicy.modelContextWindows ? { modelContextWindows: { ...staticPolicy.modelContextWindows } } : {}),
264
267
  ...(staticPolicy.modelDisplayNames ? { modelDisplayNames: { ...staticPolicy.modelDisplayNames } } : {}),
265
268
  ...(staticPolicy.modelInputModalities ? { modelInputModalities: cloneRecordOfArrays(staticPolicy.modelInputModalities) } : {}),
269
+ ...(staticPolicy.modelCapabilities ? { modelCapabilities: structuredClone(staticPolicy.modelCapabilities) } : {}),
266
270
  ...(staticPolicy.modelMaxInputTokens ? { modelMaxInputTokens: { ...staticPolicy.modelMaxInputTokens } } : {}),
267
271
  ...(entry.defaultMaxOutputTokens !== undefined ? { defaultMaxOutputTokens: entry.defaultMaxOutputTokens } : {}),
268
272
  ...(staticPolicy.modelMaxOutputTokens ? { modelMaxOutputTokens: { ...staticPolicy.modelMaxOutputTokens } } : {}),
@@ -340,6 +344,7 @@ export function deriveKeyLoginMap(): Record<string, DerivedKeyLoginProvider> {
340
344
  ...(entry.modelMaxOutputTokens ? { modelMaxOutputTokens: { ...entry.modelMaxOutputTokens } } : {}),
341
345
  ...(entry.reasoningEfforts ? { reasoningEfforts: [...entry.reasoningEfforts] } : {}),
342
346
  ...(entry.modelReasoningEfforts ? { modelReasoningEfforts: cloneRecordOfArrays(entry.modelReasoningEfforts) } : {}),
347
+ ...(entry.modelCapabilities ? { modelCapabilities: structuredClone(entry.modelCapabilities) } : {}),
343
348
  ...(entry.modelDefaultReasoningEfforts ? { modelDefaultReasoningEfforts: { ...entry.modelDefaultReasoningEfforts } } : {}),
344
349
  ...(entry.reasoningEffortMap ? { reasoningEffortMap: { ...entry.reasoningEffortMap } } : {}),
345
350
  ...(entry.modelReasoningEffortMap ? { modelReasoningEffortMap: cloneNestedRecord(entry.modelReasoningEffortMap) } : {}),
@@ -484,6 +489,8 @@ function applyVerbosityDefaults(prov: OcxProviderConfig, entry: ProviderRegistry
484
489
  */
485
490
  function enrichReasoningMetadataByDestination(prov: OcxProviderConfig): void {
486
491
  const destination = registryEntryForProviderDestination(prov);
492
+ const capabilities = mergeReasoningModelCapabilities(destination?.modelCapabilities, prov.modelCapabilities);
493
+ if (capabilities) prov.modelCapabilities = capabilities;
487
494
  applyReasoningSummaryDefaults(prov, destination?.modelSupportsReasoningSummaries);
488
495
  if (destination?.modelDefaultReasoningEfforts) {
489
496
  prov.modelDefaultReasoningEfforts = mapFill(destination.modelDefaultReasoningEfforts, prov.modelDefaultReasoningEfforts)[0];
@@ -500,6 +507,9 @@ function enrichReasoningMetadataByDestination(prov: OcxProviderConfig): void {
500
507
  if (destination?.modelReasoningEfforts) {
501
508
  prov.modelReasoningEfforts = fillRecordOfArrays(destination.modelReasoningEfforts, prov.modelReasoningEfforts);
502
509
  }
510
+ if (destination?.modelDefaultReasoningEfforts) {
511
+ [prov.modelDefaultReasoningEfforts] = mapFill(destination.modelDefaultReasoningEfforts, prov.modelDefaultReasoningEfforts);
512
+ }
503
513
  if (prov.reasoningEfforts === undefined && destination?.reasoningEfforts !== undefined) {
504
514
  prov.reasoningEfforts = [...destination.reasoningEfforts];
505
515
  }
@@ -513,6 +523,23 @@ export function hasLegacyClinePassReasoningEfforts(name: string, prov: OcxProvid
513
523
  && prov.reasoningEfforts[0] === "low";
514
524
  }
515
525
 
526
+ /**
527
+ * The registry's provider-wide `inputModalities` for a row that still points at that entry's
528
+ * destination. Runtime-only: callers apply it to a private copy (vision eligibility) or to the
529
+ * routed provider. Enrichment does not copy it, because init and management writes persist the
530
+ * enriched object, and a saved copy would follow the row to another URL.
531
+ */
532
+ export function registryInputModalitiesForDestination(name: string, prov: OcxProviderConfig): string[] | undefined {
533
+ let entry = PROVIDER_REGISTRY.find(row => row.id === name);
534
+ if (!entry || !providerMatchesRegistryTransportWithStaticGuards(name, prov)) {
535
+ entry = registryEntryForProviderDestination(prov);
536
+ // Runtime modality authority follows only the canonical fixed-key destination, not aliases.
537
+ if (entry && (entry.adapter !== prov.adapter
538
+ || normalizedProviderEndpoint(entry.baseUrl) !== normalizedProviderEndpoint(prov.baseUrl))) return undefined;
539
+ }
540
+ return entry?.inputModalities ? [...entry.inputModalities] : undefined;
541
+ }
542
+
516
543
  export function enrichProviderFromRegistry(name: string, prov: OcxProviderConfig): void {
517
544
  const entry = PROVIDER_REGISTRY.find(row => row.id === name);
518
545
  if (!entry || !providerMatchesRegistryTransportWithStaticGuards(name, prov)) {
@@ -568,6 +595,7 @@ export function enrichProviderFromRegistry(name: string, prov: OcxProviderConfig
568
595
  // (mapFill in src/providers/resolved-model-policy-merge.ts), so the wire honored the effort while /v1/models and
569
596
  // every client export showed no effort control at all.
570
597
  if (resolvedStatic.modelReasoningEfforts) prov.modelReasoningEfforts = cloneRecordOfArrays(resolvedStatic.modelReasoningEfforts);
598
+ if (resolvedStatic.modelCapabilities) prov.modelCapabilities = structuredClone(resolvedStatic.modelCapabilities);
571
599
  if (!prov.modelDefaultReasoningEfforts && seed.modelDefaultReasoningEfforts) prov.modelDefaultReasoningEfforts = { ...seed.modelDefaultReasoningEfforts };
572
600
  if (!prov.reasoningEffortMap && seed.reasoningEffortMap) prov.reasoningEffortMap = { ...seed.reasoningEffortMap };
573
601
  if (!prov.modelReasoningEffortMap && seed.modelReasoningEffortMap) prov.modelReasoningEffortMap = cloneNestedRecord(seed.modelReasoningEffortMap);
@@ -10,6 +10,8 @@ export const KIRO_MODELS = [
10
10
  // 260930 preemptive: GPT-6.1 Sol (OpenAI announced 2026-09-29; kiro.dev did not list it on 2026-09-30).
11
11
  // Mirrors gpt-6-sol; calls fail upstream until Kiro ships the model.
12
12
  "gpt-6.1-sol",
13
+ // 261027 preemptive: GPT-6.1 Terra (mirrors Kiro's GPT-6.1 Sol seed; provider listing unverified).
14
+ "gpt-6.1-terra",
13
15
  // 260929 preemptive: Claude Sonnet 5.5 added ahead of Kiro's catalog (kiro.dev did not list it on
14
16
  // 2026-09-29). Mirrors claude-sonnet-5; calls fail upstream until Kiro ships the model.
15
17
  "claude-sonnet-5.5",
@@ -46,6 +48,7 @@ export const KIRO_MODEL_CONTEXT_WINDOWS: Record<string, number> = {
46
48
  "gpt-6-sol": 272_000,
47
49
  "gpt-6-luna": 272_000,
48
50
  "gpt-6.1-sol": 272_000,
51
+ "gpt-6.1-terra": 272_000,
49
52
  "claude-sonnet-5.5": 1_000_000,
50
53
  "claude-sonnet-5": 1_000_000,
51
54
  "claude-opus-5.5": 1_000_000,
@@ -0,0 +1,12 @@
1
+ import type { ModelCapabilities } from "../types";
2
+
3
+ // Public MiMo V2.6 exposes an on/off switch, not distinct reasoning strengths.
4
+ export const MIMO_THINKING_MODELS = ["mimo-v2.6-pro", "mimo-v2.6-flash", "mimo-v2.6-pro-ultraspeed"] as const;
5
+ export const MIMO_THINKING_CAPABILITIES: Record<string, ModelCapabilities> = Object.fromEntries(
6
+ MIMO_THINKING_MODELS.map(id => [id, {
7
+ supportsReasoning: true,
8
+ supportsThinkingToggle: true,
9
+ supportsReasoningEffort: false,
10
+ defaultThinkingEnabled: true,
11
+ }]),
12
+ );
@@ -67,6 +67,7 @@ export const PROVIDER_MODEL_RENAME_ROLES = {
67
67
  modelContextWindows: "record",
68
68
  modelContextTiers: "record",
69
69
  modelInputModalities: "record",
70
+ inputModalities: "none",
70
71
  modelCapabilities: "record",
71
72
  modelMaxInputTokens: "record",
72
73
  modelAutoCompactTokenLimits: "record",
@@ -19,6 +19,7 @@
19
19
  // Everything else in the row is left alone.
20
20
 
21
21
  import { PROVIDER_REGISTRY, normalizedProviderEndpoint } from "./registry";
22
+ import { MINIMAX_M31_FLASH, MINIMAX_M31_FLASH_PREVIEW_RETIRED } from "./registry/model-seeds";
22
23
  import { MODEL_KEYED_RECORDS, MODEL_ID_LISTS, MODEL_ID_SCALARS, MODEL_NESTED_RECORDS } from "./model-rename-fields";
23
24
  import { providerConfigSeed } from "./derive";
24
25
  import type { OcxConfig, OcxProviderConfig } from "../types";
@@ -138,6 +139,15 @@ export const MODEL_RENAMES: readonly ModelRename[] = [
138
139
  // may also hold one as a value. 3.8 resolves tiers from its own suffix ladder instead.
139
140
  dropReasoningEffortMap: true,
140
141
  })),
142
+ // MiniMax shipped M3.1-Flash as generally available on 2026-10-19 and retired the preview
143
+ // id, so a saved MiniMax-M3.1-Flash-Preview selection would answer model-not-found. The
144
+ // stale-roster refresh only matches exact old seeds; this repairs the preview id itself.
145
+ ...(["minimax", "minimax-cn"] as const).map(provider => ({
146
+ provider,
147
+ from: MINIMAX_M31_FLASH_PREVIEW_RETIRED,
148
+ to: MINIMAX_M31_FLASH,
149
+ reason: "MiniMax shipped M3.1-Flash as generally available and retired the preview id",
150
+ })),
141
151
  ];
142
152
 
143
153
  function renameInList(value: unknown, from: string, to: string): string[] | null {
@@ -5,7 +5,7 @@ import {
5
5
  type CodexAuthContext,
6
6
  type CodexAuthPolicyConfig,
7
7
  } from "../codex/auth-context";
8
- import { poolContextCreditHoldResetAt } from "../codex/pool-credit-policy";
8
+ import { poolContextCreditHoldResetAt, poolContextCreditRecoveryNeeded } from "../codex/pool-credit-policy";
9
9
 
10
10
  export function createOpenAiSidecarCreditGuard(
11
11
  context: CodexAuthContext,
@@ -16,7 +16,7 @@ export function createOpenAiSidecarCreditGuard(
16
16
  const resetAt = poolContextCreditHoldResetAt(context, policy);
17
17
  if (resetAt === undefined) return;
18
18
  releaseCodexAuthContextProbeLease(context);
19
- throw new CodexPoolAccountCreditsOffError(context.accountId, resetAt);
19
+ throw new CodexPoolAccountCreditsOffError(context.accountId, resetAt, poolContextCreditRecoveryNeeded(context, policy), context.quotaScope);
20
20
  };
21
21
  }
22
22
 
@@ -0,0 +1,76 @@
1
+ import type { ModelCapabilities, OcxProviderConfig, ReasoningCapabilities } from "../types";
2
+ import { modelInList } from "../types";
3
+ import { registryEntryForProviderDestination } from "./registry";
4
+ import { legacyModelValue, mapFill, nestedMapFill } from "./resolved-model-policy-merge";
5
+
6
+ export const REASONING_CAPABILITY_FIELDS = [
7
+ "supportsReasoning", "supportsThinkingToggle", "supportsReasoningEffort", "defaultThinkingEnabled",
8
+ ] as const;
9
+
10
+ /** Case-equal model declarations merge per axis; unrelated axes retain registry defaults. */
11
+ export function mergeReasoningModelCapabilities(
12
+ registry: Record<string, ModelCapabilities> | undefined,
13
+ explicit: Record<string, ModelCapabilities> | undefined,
14
+ ): Record<string, ModelCapabilities> | undefined {
15
+ if (!registry && !explicit) return undefined;
16
+ const merged: Record<string, ModelCapabilities> = {};
17
+ const keys = new Map<string, string>();
18
+ for (const rows of [registry, explicit]) {
19
+ for (const [id, row] of Object.entries(rows ?? {})) {
20
+ const key = keys.get(id.toLowerCase()) ?? id;
21
+ keys.set(id.toLowerCase(), key);
22
+ merged[key] = { ...merged[key], ...structuredClone(row) };
23
+ }
24
+ }
25
+ return merged;
26
+ }
27
+
28
+ /** Match a declared model case-insensitively, without inheriting suffix or family aliases. */
29
+ export function declaredModelCapabilities(rows: Readonly<Record<string, ModelCapabilities>> | undefined, modelId: string): ModelCapabilities | undefined {
30
+ return rows?.[modelId] ?? Object.entries(rows ?? {}).find(([id]) => id.toLowerCase() === modelId.toLowerCase())?.[1];
31
+ }
32
+
33
+ /** Captured-policy variant: never reads the live registry after admission. */
34
+ export function reasoningCapabilitiesFromConfig(provider: Readonly<OcxProviderConfig>, modelId: string): ReasoningCapabilities {
35
+ if (modelInList(provider.noReasoningModels, modelId)) {
36
+ return { supportsReasoning: false, supportsThinkingToggle: false, supportsReasoningEffort: false, defaultThinkingEnabled: false };
37
+ }
38
+ const declared = declaredModelCapabilities(provider.modelCapabilities, modelId);
39
+ const result: ReasoningCapabilities = {};
40
+ for (const field of REASONING_CAPABILITY_FIELDS) {
41
+ if (typeof declared?.[field] === "boolean") result[field] = declared[field];
42
+ }
43
+ const efforts = legacyModelValue(provider.modelReasoningEfforts, modelId) ?? provider.reasoningEfforts;
44
+ const effortMap = legacyModelValue(provider.modelReasoningEffortMap, modelId) ?? provider.reasoningEffortMap;
45
+ const effectiveEfforts = efforts?.filter(effort => effort !== "none").map(effort => effortMap?.[effort] ?? effort);
46
+ const strengths = ["minimal", "low", "medium", "high", "xhigh", "max", "ultra"];
47
+ if (result.supportsReasoningEffort === undefined && effectiveEfforts?.some(effort => strengths.includes(effort))) {
48
+ result.supportsReasoningEffort = true;
49
+ }
50
+ if (result.supportsReasoning === undefined && (result.supportsThinkingToggle === true
51
+ || result.supportsReasoningEffort === true || effectiveEfforts?.some(effort => strengths.includes(effort) || effort === "enabled" || effort === "adaptive")
52
+ || modelInList(provider.thinkingBudgetModels, modelId))) {
53
+ result.supportsReasoning = true;
54
+ }
55
+ if (result.supportsReasoning === false) {
56
+ result.supportsThinkingToggle = false;
57
+ result.supportsReasoningEffort = false;
58
+ result.defaultThinkingEnabled = false;
59
+ }
60
+ return result;
61
+ }
62
+
63
+ /** Resolve fixed destination registry facts beneath exact operator capabilities. Unknown stays absent. */
64
+ export function resolveReasoningCapabilities(provider: OcxProviderConfig, modelId: string): ReasoningCapabilities {
65
+ const entry = registryEntryForProviderDestination(provider);
66
+ return reasoningCapabilitiesFromConfig({
67
+ ...provider,
68
+ modelCapabilities: mergeReasoningModelCapabilities(entry?.modelCapabilities, provider.modelCapabilities),
69
+ modelReasoningEfforts: mapFill(entry?.modelReasoningEfforts, provider.modelReasoningEfforts)[0],
70
+ reasoningEfforts: provider.reasoningEfforts ?? entry?.reasoningEfforts,
71
+ reasoningEffortMap: mapFill(entry?.reasoningEffortMap, provider.reasoningEffortMap)[0],
72
+ modelReasoningEffortMap: nestedMapFill(entry?.modelReasoningEffortMap, provider.modelReasoningEffortMap)[0],
73
+ thinkingBudgetModels: [...(entry?.thinkingBudgetModels ?? []), ...(provider.thinkingBudgetModels ?? [])],
74
+ noReasoningModels: [...(entry?.noReasoningModels ?? []), ...(provider.noReasoningModels ?? [])],
75
+ }, modelId);
76
+ }
@@ -251,7 +251,8 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
251
251
  // (it is the current catalog, so its default ordering wins), then the ids
252
252
  // only the old devin entry carried. Degraded-mode seed only either way —
253
253
  // `liveModels` discovers the account's real roster.
254
- models: ["swe-2", "swe-1-7", "gpt-5-6-sol", "gpt-6-astra", "gpt-6-1-sol", "gpt-6-sol", "gpt-6-luna", "claude-opus-5-5", "claude-opus-5", "claude-fable-5-1", "claude-sonnet-5-5", "claude-haiku-5-5", "claude-sonnet-5", "glm-5-3", "kimi-k3", "gemini-3-8-flash", "grok-4-6", "grok-4-7", "swe-1-7-lightning", "gpt-5-6-luna", "gpt-5-6-terra", "claude-opus-4-8", "glm-5-2", "kimi-k2-7", "grok-4-5"],
254
+ // 261027 preemptive: GPT-6.1 Terra (same degraded seed policy as GPT-6.1 Sol).
255
+ models: ["swe-2", "swe-1-7", "gpt-5-6-sol", "gpt-6-astra", "gpt-6-1-sol", "gpt-6-1-terra", "gpt-6-sol", "gpt-6-luna", "claude-opus-5-5", "claude-opus-5", "claude-fable-5-1", "claude-sonnet-5-5", "claude-haiku-5-5", "claude-sonnet-5", "glm-5-3", "kimi-k3", "gemini-3-8-flash", "grok-4-6", "grok-4-7", "swe-1-7-lightning", "gpt-5-6-luna", "gpt-5-6-terra", "claude-opus-4-8", "glm-5-2", "kimi-k2-7", "grok-4-5"],
255
256
  liveModels: true,
256
257
  defaultModel: "swe-2",
257
258
  modelContextWindows: DEVIN_MODEL_CONTEXT_WINDOWS,
@@ -757,6 +758,8 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
757
758
  ...OPENAI_DAYBREAK_REASONING_EFFORTS,
758
759
  "gpt-6-astra": ["low", "medium", "high", "xhigh", "max"],
759
760
  ...Object.fromEntries(OPENAI_GPT6_MODELS.map(id => [id, ["low", "medium", "high", "xhigh", "max"]])),
761
+ // OpenAI API model page (2026-10-27): Terra also accepts none; Sol does not.
762
+ "gpt-6.1-terra": ["none", "low", "medium", "high", "xhigh", "max"],
760
763
  },
761
764
  virtualModels: OPENAI_API_GPT56_VIRTUAL_MODELS,
762
765
  },
@@ -1098,6 +1101,7 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
1098
1101
  "openai/gpt-6-luna": true,
1099
1102
  // Listed 2026-09-29 as an OpenAI-backed route like GPT-6 Sol.
1100
1103
  "openai/gpt-6.1-sol": true,
1104
+ "openai/gpt-6.1-terra": true,
1101
1105
  },
1102
1106
  // Deliberately no OpenRouter route pin: it bills the endpoint actually used and reports the
1103
1107
  // actual service_tier. B0 confirmation therefore owns downgrade safety. Forcing `only` plus
@@ -1256,17 +1260,18 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
1256
1260
  // the list only as compatibility aliases so existing saved configs and requests
1257
1261
  // keep validating and routing (they previously mapped to v4-flash; devlog
1258
1262
  // _fin/260710_provider_hardening/002_research_cn.md). The current offerings are
1259
- // V4.1-Flash — defaultModel and the model-specific wiring below use its live id.
1263
+ // V4.1-Flash and V4 Pro; Flash remains the default.
1260
1264
  // Keep the legacy vision-preview alias; see DEEPSEEK_VISION_PREVIEW_MODEL.
1261
1265
  models: ["deepseek-chat", "deepseek-reasoner", ...DEEPSEEK_NATIVE_THINKING_MODELS, DEEPSEEK_VISION_PREVIEW_MODEL],
1262
- // V4.1-Flash is the current first-party offering; `deepseek-v4-flash` now routes there
1266
+ // V4.1-Flash is the first-party default; `deepseek-v4-flash` now routes there
1263
1267
  // as a compatibility alias, so a new install should ask for the live id by name.
1264
1268
  defaultModel: "deepseek-flash",
1265
1269
  // Official DeepSeek Codex setup (codex-deepseek-setup.sh) advertises 1,048,576
1266
1270
  // for both V4 models; the older 1,000,000 figure was a rounded approximation.
1267
- modelContextWindows: { "deepseek-flash": 1_048_576, "deepseek-v4-flash": 1_048_576, [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576 },
1271
+ modelContextWindows: { "deepseek-flash": 1_048_576, "deepseek-v4-flash": 1_048_576, "deepseek-v4-pro": 1_048_576, [DEEPSEEK_VISION_PREVIEW_MODEL]: 1_048_576 },
1268
1272
  modelInputModalities: {
1269
1273
  "deepseek-flash": ["text", "image"],
1274
+ "deepseek-v4-pro": ["text"],
1270
1275
  [DEEPSEEK_VISION_PREVIEW_MODEL]: ["text", "image"],
1271
1276
  },
1272
1277
  // DeepSeek documents both V4 models as native Responses API models adapted for Codex
@@ -1281,6 +1286,7 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
1281
1286
  // translating them into Responses would add a hop onto our newest upstream path
1282
1287
  // for no gain.
1283
1288
  "deepseek-v4-flash": { wire: "openai-responses", inbound: ["responses"] },
1289
+ "deepseek-v4-pro": { wire: "openai-responses", inbound: ["responses"] },
1284
1290
  // Same Responses contract as the V4 ids it succeeds; without this row the new
1285
1291
  // default would fall back to the provider-wide Chat wire.
1286
1292
  "deepseek-flash": { wire: "openai-responses", inbound: ["responses"] },
@@ -1298,7 +1304,7 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
1298
1304
  // devlog/_fin/260807_deepseek_responses_streaming/000_plan.md.
1299
1305
  // Current official streams normally carry a real terminal; retain a narrow grace
1300
1306
  // repair for the historical shape that closes after a complete graph without one.
1301
- modelResponsesTerminalRepair: { "deepseek-flash": { graceMs: 5_000 }, "deepseek-v4-flash": { graceMs: 5_000 } },
1307
+ modelResponsesTerminalRepair: { "deepseek-flash": { graceMs: 5_000 }, "deepseek-v4-flash": { graceMs: 5_000 }, "deepseek-v4-pro": { graceMs: 5_000 } },
1302
1308
  // DeepSeek's Responses route emits bare UUID item ids, which leave Codex
1303
1309
  // clients stuck on an uncommitted turn (#938). Client-facing only — raw
1304
1310
  // continuation snapshots keep the upstream ids.
@@ -1337,16 +1343,12 @@ export const PROVIDER_REGISTRY_CORE: readonly ProviderRegistryEntry[] = [
1337
1343
  modelReasoningEfforts: Object.fromEntries(DEEPSEEK_NATIVE_THINKING_MODELS.map(id => [id, deepseekThinkingEffortsFor(id)])),
1338
1344
  modelReasoningEffortMap: Object.fromEntries(DEEPSEEK_NATIVE_THINKING_MODELS.map(id => [id, deepseekReasoningMapFor(id)])),
1339
1345
  modelSupportsReasoningSummaries: Object.fromEntries(DEEPSEEK_NATIVE_THINKING_MODELS.map(id => [id, true])),
1340
- // deepseek-v4-pro is absent from the live sets on purpose (retired upstream, requests
1341
- // route to V4.1-Flash — see model-seeds.ts), but it stays SELECTABLE in configs that
1342
- // already carry it, and what serves it is a thinking-mode model. A saved config with
1343
- // the alias selected would otherwise serialize a bare continuation the upstream 400s
1344
- // on (#5421). The preserve list governs replay, not advertisement.
1345
- preserveReasoningContentModels: [...DEEPSEEK_NATIVE_THINKING_MODELS, "deepseek-v4-pro"],
1346
+ // Native thinking models retain assistant reasoning on tool-call continuations (#5421).
1347
+ preserveReasoningContentModels: DEEPSEEK_NATIVE_THINKING_MODELS,
1346
1348
  // #4436: first-party deepseek-flash accepts native images on Chat and Responses.
1347
1349
  // Keep unprobed compatibility aliases on the #88 sidecar path. This must be fixed
1348
1350
  // here: router enrichment unions this list with saved config, so config cannot remove it.
1349
- noVisionModels: ["deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash"],
1351
+ noVisionModels: ["deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash", "deepseek-v4-pro"],
1350
1352
  },
1351
1353
  // llama-3.3-70b was deprecated by Cerebras on 2026-02-16. Evidence: devlog/_plan/260710_provider_hardening/003_research_aggregators.md.
1352
1354
  { id: "cerebras", label: "Cerebras", baseUrl: "https://api.cerebras.ai/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://cloud.cerebras.ai/platform/apikeys", defaultModel: "gpt-oss-120b" },