@oh-my-pi/pi-catalog 18.4.2 → 18.4.4

package/CHANGELOG.md

@@ -2,6 +2,49 @@
 
 ## [Unreleased]
 
+## [18.4.4] - 2026-09-29
+
+### Added
+
+- Added `compat.bedrockMessagesApi` for `anthropic-messages` models: detected from a Bedrock `/anthropic` base URL under any provider id, it drops tool `strict`, fits `metadata.user_id` to Bedrock's pattern, and enables on-demand compaction; set it in `models.yml` to opt a proxy or an `ANTHROPIC_BASE_URL` reroute in, or `false` to opt out ([#13311](https://github.com/can1357/oh-my-pi/pull/13311)).
+- Added GPT-6.1 Sol pricing for `openai-codex` (`gpt-6.1-sol`, `gpt-6.1-sol-wm`: $2 input, $10 output, $0.10 cached input), so Codex usage shows cost instead of $0 ([#13782](https://github.com/can1357/oh-my-pi/pull/13782) by [@H4vC](https://github.com/H4vC)).
+- Added `Model.serviceTiers`, the service tiers a provider advertises for a model; Codex discovery fills it from `service_tiers` (e.g. `priority`, `ultrafast`) ([#13782](https://github.com/can1357/oh-my-pi/pull/13782) by [@H4vC](https://github.com/H4vC)).
+- Added the documented 922K input maximum for `openai-codex/gpt-6.1-sol` with extended context on (Codex reports a stale 872K), matching GPT-6 Astra; the default window stays 272K ([#13782](https://github.com/can1357/oh-my-pi/pull/13782) by [@H4vC](https://github.com/H4vC)).
+
+### Fixed
+
+- Fixed `openai-codex/gpt-6.1-sol` not appearing in Codex discovery even on accounts where the Codex CLI lists it: the backend hides it from client version 0.155.1, so Codex requests now report 0.159.0, the current Codex CLI release ([#13782](https://github.com/can1357/oh-my-pi/pull/13782) by [@H4vC](https://github.com/H4vC)).
+- Fixed on-demand compaction staying off for Claude models on the `amazon-bedrock` and `bedrock-mantle` providers when they use Bedrock's `/anthropic` routes ([#13311](https://github.com/can1357/oh-my-pi/pull/13311) by [@mustafaabidali](https://github.com/mustafaabidali)).
+- Fixed Claude models on Bedrock's `/anthropic` routes resolving `compat.disableStrictTools: false`, although those routes reject the tool `strict` field ([#13311](https://github.com/can1357/oh-my-pi/pull/13311) by [@mustafaabidali](https://github.com/mustafaabidali)).
+- Fixed Bedrock's FIPS (`bedrock-runtime-fips`) and AWS PrivateLink (`vpce-….vpce.amazonaws.com`) hostnames, and Mantle's documented `/v1` OpenAI base, not being recognized as Bedrock routes, which left them without native compaction and the `/anthropic` request fixes ([#13311](https://github.com/can1357/oh-my-pi/pull/13311) by [@mustafaabidali](https://github.com/mustafaabidali)).
+- Added `supportsBetweenToolsThinking` Anthropic compat flag (`supports-between-tools-thinking` KDL axis), enabled for Claude Sonnet 5.5
+
+## [18.4.3] - 2026-09-28
+
+### Added
+
+- Added Command Code's typesafe/jev decision model for the judge role ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Added Command Code's DeepSeek V4.1 Flash Fast model ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Added Command Code's Claude Sonnet 5.5 model ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Added `trust-forbidden=#true` for optional login `validate` checks, so a key check that answers 403 keeps the pasted key instead of rejecting it ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- `hosted-image #false` and `image-model #false` now remove a hosted image flag or image model that a class rule grants ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Added support for Claude Sonnet 5.5 model with image and text inputs
+- Added new compatibility rules for Anthropic Sonnet family enabling mid‑conversation system features and disabling forced tool choice
+- Added the `web-search-model`, `hosted-image`, and `image-model` catalog axes (`Model.webSearchModel`, `hostedImage`, `imageModel`). `web-search` now comes from the model's lineage and API (GPT-5+ Responses, Claude 4+ Messages, Gemini 2+), so proxies and gateways that expose these models inherit it.
+
+### Changed
+
+- Renamed the Codex image model `openai-codex/gpt-image-1` to `openai-codex/gpt-image-2` to match what the Codex backend runs (`gpt-image-2-codex`); saved `openai-codex/gpt-image-1` selectors resolve to the new id.
+- Routed Command Code's 10 GPT models through the OpenAI Responses API ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Command Code login now rejects a key that Command Code answers with 401; a 403 or an unreachable check keeps the pasted key, as the Command Code CLI does ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Command Code GPT models no longer advertise hosted image generation or the `gpt-image-2` image model, which Command Code does not serve ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Command Code Muse Spark models drop the minimal thinking level, and Muse Spark 1.3 adds max, matching the Command Code CLI ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+- Changed Command Code prices to match the Command Code CLI: DeepSeek V4 Flash and V4 Flash Vision Exp drop from 0.22 / 0.66 to 0.15 / 0.6 per 1M tokens, Step 3.5 Flash input drops from 0.1 to 0.09, and LongCat 2.0 is no longer shown as free (0.3 / 1.2) ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+
+### Fixed
+
+- Fixed missing thinking levels, image input, and prices for Command Code models ([#13666](https://github.com/can1357/oh-my-pi/pull/13666) by [@riicodespretty](https://github.com/riicodespretty))
+
 ## [18.4.2] - 2026-09-28
 
 ### Changed

package/dist/types/compat/anthropic.d.ts

@@ -6,6 +6,12 @@
  * prefix check would accept lookalikes like `https://api.anthropic.com.evil.com`.
  */
 export declare function isOfficialAnthropicApiUrl(baseUrl?: string): boolean;
+/**
+ * Amazon Bedrock's Anthropic Messages API: the `/anthropic` path on the
+ * bedrock-runtime or bedrock-mantle endpoint. It rejects the tool `strict`
+ * field and enforces Anthropic signatures on replay.
+ */
+export declare function isBedrockAnthropicRoute(baseUrl?: string): boolean;
 /**
  * Azure AI Inference / Foundry Anthropic route
  * (`<resource>.inference.ai.azure.com`, `<resource>.services.ai.azure.com`).

package/dist/types/compat/types.d.ts

@@ -350,12 +350,15 @@ export type CompiledAuthValidation = {
     maxTokensField?: "max_tokens" | "max_completion_tokens";
     maxTokens?: number;
     optional?: boolean;
+    /** With `optional`: a 403 also trusts the key; only a 401 rejects it. */
+    trustForbidden?: boolean;
 } | {
     kind: "anthropic-messages";
     label?: string;
     baseUrl: string;
     model: string;
     optional?: boolean;
+    trustForbidden?: boolean;
 } | {
     kind: "models-endpoint";
     label?: string;
@@ -365,6 +368,7 @@ export type CompiledAuthValidation = {
     /** Hook returning extra request headers (may throw a configuration error). */
     headersHook?: string;
     optional?: boolean;
+    trustForbidden?: boolean;
 };
 /** Paste-an-API-key login: optional browser hint, prompt, optional validation. */
 export interface CompiledApiKeyLogin {

package/dist/types/hosts.d.ts

@@ -165,5 +165,18 @@ export declare function isVertexExpressOpenAIUrl(baseUrl: string): boolean;
 export declare function isVertexRawPredictUrl(baseUrl: string): boolean;
 /** Azure OpenAI deployment-scoped path (`…/deployments/<name>/…`). */
 export declare function isAzureDeploymentsUrl(baseUrl: string): boolean;
+/**
+ * Amazon Bedrock API route on a bedrock-runtime or bedrock-mantle endpoint
+ * (public, FIPS, or PrivateLink hostname):
+ * - `anthropic`: the Anthropic Messages API under `/anthropic`.
+ * - `openai`: the OpenAI-compatible APIs under `/openai`, plus Mantle's
+ *   documented `/v1` base (`https://bedrock-mantle.<region>.api.aws/v1`).
+ *
+ * Hostnames are parsed strictly so proxies that embed these hosts in a path
+ * do not match.
+ */
+export declare function isBedrockRouteUrl(baseUrl: string | undefined, route: "openai" | "anthropic"): boolean;
+/** Amazon Bedrock's OpenAI-compatible routes; see {@link isBedrockRouteUrl}. */
+export declare function isBedrockOpenAIUrl(baseUrl: string | undefined): boolean;
 /** Alibaba DashScope consumer `compatible-mode` endpoint (rejects multimodal arrays for some text-only SKUs). */
 export declare function isDashscopeCompatibleModeUrl(baseUrl: string): boolean;

package/dist/types/provider-models/openai-compat.d.ts

@@ -648,7 +648,8 @@ export declare function modelsDevCatalogFallback(providerId: string, fetchImpl?:
  * `baseUrl` overrides the Provider API base path for testing; it is
  * normalized to the shared `/provider` root (a trailing `/v1` is stripped)
  * so Claude ids route to the Anthropic-compatible Messages endpoint at the
- * root while every other id uses chat completions under `/v1`.
+ * root, the ten GPT ids listed in the `api-routes` table to the Responses
+ * endpoint under `/v1`, and every other id to chat completions under `/v1`.
  */
 export interface CommandCodeModelManagerConfig {
     apiKey?: string;
@@ -659,12 +660,14 @@ export interface CommandCodeModelManagerConfig {
  * Builds the Command Code model manager: a mixed-protocol OpenAI-compatible
  * discovery client. The public `/v1/models` catalog is fetched once per
  * options instance; `mapModel` pins each row's transport from the
- * `api-routes` table (Claude ids to `anthropic-messages`, everything else to
- * `openai-completions`) and seeds neutral capability defaults. Reviewed
- * Command Code policy (effort ladders, pricing, limits, modalities) is
- * applied later by `buildModel` from `providers/commandcode.kdl` — the
- * mapper never inherits another provider's reasoning, rates, image support,
- * or context window.
+ * `api-routes` table (Claude ids to `anthropic-messages`, the listed GPT ids
+ * to `openai-responses`, everything else to `openai-completions`) and seeds
+ * neutral capability defaults. Reviewed Command Code policy (effort ladders,
+ * pricing, limits, modalities) is applied later by `buildModel` from
+ * `providers/commandcode.kdl` — the mapper never inherits another provider's
+ * reasoning, rates, image support, or context window. A successful fetch also
+ * appends the KDL `seed` rows (typesafe/jev), rebased onto the configured
+ * base, because `dynamicModelsAuthoritative` would prune `staticModels`.
  */
 export declare function commandCodeModelManagerOptions(config?: CommandCodeModelManagerConfig): ModelManagerOptions<Api>;
 /**

package/dist/types/types.d.ts

@@ -481,6 +481,15 @@ export interface AnthropicCompat {
      * `tools.<n>.custom.strict: Extra inputs are not permitted`.
      */
     disableStrictTools?: boolean;
+    /**
+     * The endpoint is Amazon Bedrock's Anthropic Messages API (`/anthropic` on
+     * bedrock-runtime or bedrock-mantle). Requests drop tool `strict`, fit
+     * `metadata.user_id` to Bedrock's request-metadata pattern, and may use
+     * on-demand compaction. Detected from the model `baseUrl`; set it in
+     * models.yml `compat` for a route the URL check cannot see (a proxy, an
+     * `ANTHROPIC_BASE_URL` reroute) or to `false` to opt out.
+     */
+    bedrockMessagesApi?: boolean;
     /**
      * Map adaptive thinking (`thinking: { type: "adaptive" }`) to
      * `{ type: "enabled", budget_tokens }`. Vertex AI rejects the `adaptive`
@@ -508,6 +517,13 @@ export interface AnthropicCompat {
      * `input_transformations` under the thinking-binding-controls beta.
      */
     supportsThinkingBindingControls?: boolean;
+    /**
+     * Whether the model replaces `thinking: { type: "disabled" }` with
+     * `thinking: { type: "between_tools" }` (Claude Sonnet 5.5). The disabled
+     * form is rejected with a 400; `between_tools` skips up-front thinking and
+     * only allows progress updates between tool calls.
+     */
+    supportsBetweenToolsThinking?: boolean;
     /**
      * Whether the model accepts a forced `tool_choice` (`{ type: "any" }` or
      * `{ type: "tool", name }`). Claude Fable/Mythos 5 reject forced tool use
@@ -804,9 +820,15 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
  */
 export type ResolvedOpenRouterCompat = ResolvedOpenAICompat & ResolvedOpenAIResponsesCompat;
 /** Fully-resolved anthropic-messages compat view (same contract as `ResolvedOpenAICompat`). */
-export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard">> & {
+export type ResolvedAnthropicCompat = Required<Omit<AnthropicCompat, "streamIdleTimeoutMs" | "thinkingLoopGuard" | "bedrockMessagesApi">> & {
     /** Thinking-loop watchdog guard family applied to streamed reasoning. */
     thinkingLoopGuard?: AnthropicCompat["thinkingLoopGuard"];
+    /**
+     * Bedrock's Anthropic Messages API (see {@link AnthropicCompat.bedrockMessagesApi}).
+     * `true` when detected or configured; otherwise unset, so rows baked before
+     * this field keep matching.
+     */
+    bedrockMessagesApi?: boolean;
     /**
      * Stream-watchdog idle-timeout fallback in ms for slow reasoning hosts; 0 disables the idle watchdog.
      * Undefined defers to `PI_STREAM_IDLE_TIMEOUT_MS`, then the legacy
@@ -986,6 +1008,12 @@ export interface Model<TApi extends Api = Api> {
     kind?: ModelKind;
     /** Grounding transport supported by this chat model. */
     webSearch?: WebSearchGrounding;
+    /** Cheaper same-provider model to run hosted web search in this model's place (model id or provider/id). */
+    webSearchModel?: string;
+    /** Whether this chat model can carry the Responses `image_generation` tool itself. */
+    hostedImage?: boolean;
+    /** Same-provider image model to generate images in this model's place (model id or provider/id). */
+    imageModel?: string;
     /**
      * Structured model identity resolved by the compat engine: vendor lineage
      * class, product family, and revision. Baked into models.json rows and
@@ -1137,6 +1165,12 @@ export interface Model<TApi extends Api = Api> {
     useResponsesLite?: boolean;
     /** Codex Code Mode restriction: model expects tools routed through a programmatic exec surface (mirrors codex-rs `tool_mode`). */
     toolMode?: "code_mode_only";
+    /**
+     * Service-tier ids the provider advertises for this model (Codex discovery
+     * `service_tiers[].id`, e.g. `priority`, `ultrafast`). Absent when the
+     * provider publishes no per-model tier list.
+     */
+    serviceTiers?: readonly string[];
     /** Preferred model to switch to when context promotion is triggered (model id or provider/id). */
     contextPromotionTarget?: string;
     /** Preferred model to use only for compaction (model id or provider/id); the active session model is unchanged. */

package/dist/types/wire/codex.d.ts

@@ -7,10 +7,12 @@ export declare const CODEX_BASE_URL = "https://chatgpt.com/backend-api";
  * Updated from npm by `bun run check-spoofed-versions --update`.
  *
  * The backend version-gates model availability against this value on both
- * `/models?client_version=` and `/responses`; 0.155.1 exposes GPT-6 Sol and Luna.
- * An older pin silently hides newer SKUs from discovery.
+ * `/models?client_version=` and `/responses`, independently of a model's
+ * declared `minimal_client_version`: GPT-6.1 Sol declares 0.153.0 yet is
+ * omitted from `/models` at 0.155.1 and listed at 0.159.0. An older pin
+ * silently hides newer SKUs from discovery.
  */
-export declare const CODEX_CLIENT_VERSION = "0.155.1";
+export declare const CODEX_CLIENT_VERSION = "0.159.0";
 export declare const OPENAI_HEADERS: {
     readonly BETA: "OpenAI-Beta";
     /** Codex feature-negotiation header; values identify opt-in wire protocols. */

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.4.2",
+	"version": "18.4.4",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.4.2",
-		"@oh-my-pi/pi-utils": "18.4.2"
+		"@oh-my-pi/omptype": "18.4.4",
+		"@oh-my-pi/pi-utils": "18.4.4"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.4.2",
+		"@oh-my-pi/pi-ai": "18.4.4",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/build.ts

@@ -87,6 +87,11 @@ function applyCatalogAssignments<TApi extends Api>(model: Model<TApi>, catalog:
 	) {
 		model.webSearch = webSearch;
 	}
+	if (typeof catalog.webSearchModel === "string") model.webSearchModel = catalog.webSearchModel;
+	if (catalog.hostedImage === true) model.hostedImage = true;
+	else if (catalog.hostedImage === false) delete model.hostedImage;
+	if (typeof catalog.imageModel === "string") model.imageModel = catalog.imageModel;
+	else if (catalog.imageModel === false) delete model.imageModel;
 	const serviceTierCost = objectPayload(catalog.serviceTierCost);
 	if (serviceTierCost !== undefined) {
 		const flex = numberField(serviceTierCost, "flex");

package/src/compat/anthropic.ts

@@ -5,7 +5,7 @@
  * pi-ai both consume these; URL matching is the one detection surface that
  * stays in code.
  */
-import { hostMatchesUrl } from "../hosts";
+import { hostMatchesUrl, isBedrockRouteUrl } from "../hosts";
 
 const OFFICIAL_ANTHROPIC_URL = "https://api.anthropic.com";
 
@@ -27,6 +27,15 @@ const VERTEX_ANTHROPIC_URL_MARKER = /aiplatform\.googleapis\.com\/.+\/publishers
 const BEDROCK_ANTHROPIC_URL_MARKER = /(?:^|\/\/|\.)bedrock-runtime\.[a-z0-9-]+\.amazonaws\.com/i;
 const AZURE_ANTHROPIC_URL_MARKER = /(?:^|\/\/|\.)[a-z0-9-]+\.(?:inference|services)\.ai\.azure\.com/i;
 
+/**
+ * Amazon Bedrock's Anthropic Messages API: the `/anthropic` path on the
+ * bedrock-runtime or bedrock-mantle endpoint. It rejects the tool `strict`
+ * field and enforces Anthropic signatures on replay.
+ */
+export function isBedrockAnthropicRoute(baseUrl?: string): boolean {
+	return isBedrockRouteUrl(baseUrl, "anthropic");
+}
+
 /**
  * Azure AI Inference / Foundry Anthropic route
  * (`<resource>.inference.ai.azure.com`, `<resource>.services.ai.azure.com`).
@@ -53,6 +62,7 @@ export function isAnthropicSigningProxyUrl(baseUrl?: string): boolean {
 			(CLOUDFLARE_ANTHROPIC_GATEWAY_URL_MARKER.test(baseUrl) ||
 				VERTEX_ANTHROPIC_URL_MARKER.test(baseUrl) ||
 				BEDROCK_ANTHROPIC_URL_MARKER.test(baseUrl))) ||
+		isBedrockAnthropicRoute(baseUrl) ||
 		isAzureAnthropicRoute(baseUrl)
 	);
 }

package/src/compat/axes.ts

@@ -190,6 +190,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"zai-reasoning-effort-dialect": wire("zaiReasoningEffortDialect", ["openai"]),
 
 	// ── wire: anthropic-messages ──
+	"bedrock-messages-api": wire("bedrockMessagesApi", ["anthropic"]),
 	"allow-anthropic-header-overrides": wire("allowAnthropicHeaderOverrides", ["anthropic"]),
 	"disable-adaptive-thinking": wire("disableAdaptiveThinking", ["anthropic"]),
 	"disable-strict-tools": wire("disableStrictTools", ["anthropic"]),
@@ -209,6 +210,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"supports-mid-conversation-tool-changes": wire("supportsMidConversationToolChanges", ["anthropic"]),
 	"supports-per-message-effort": wire("supportsPerMessageEffort", ["anthropic"]),
 	"supports-server-compaction": wire("supportsServerCompaction", ["anthropic"]),
+	"supports-between-tools-thinking": wire("supportsBetweenToolsThinking", ["anthropic"]),
 	"supports-thinking-binding-controls": wire("supportsThinkingBindingControls", ["anthropic"]),
 	"supports-turn-scoped-system": wire("supportsTurnScopedSystem", ["anthropic"]),
 
@@ -343,6 +345,9 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		shape: "scalar",
 		values: ["gemini", "anthropic", "codex", "xai", "openrouter", "openai"],
 	},
+	"web-search-model": { key: "webSearchModel", set: "catalog", shape: "scalar" },
+	"hosted-image": { key: "hostedImage", set: "catalog", shape: "scalar", values: [true, false] },
+	"image-model": { key: "imageModel", set: "catalog", shape: "scalar" },
 	"limits-patch": { key: "limitsPatch", set: "catalog", shape: "object" },
 	"long-context-cost": { key: "longContext", set: "catalog", shape: "object" },
 	"prompt-cache": { key: "promptCache", set: "catalog", shape: "object" },

package/src/compat/collapse.ts

@@ -1033,8 +1033,17 @@ function retargetCollapsedModelReferences<TSpec extends VariantSpecLike>(specs:
 			liveIdsByProvider,
 		);
 		const compactionModel = resolveCollapsedModelReference(spec.compactionModel, spec.provider, liveIdsByProvider);
-		if (contextPromotionTarget === spec.contextPromotionTarget && compactionModel === spec.compactionModel) continue;
-		specs[index] = { ...spec, contextPromotionTarget, compactionModel };
+		const webSearchModel = resolveCollapsedModelReference(spec.webSearchModel, spec.provider, liveIdsByProvider);
+		const imageModel = resolveCollapsedModelReference(spec.imageModel, spec.provider, liveIdsByProvider);
+		if (
+			contextPromotionTarget === spec.contextPromotionTarget &&
+			compactionModel === spec.compactionModel &&
+			webSearchModel === spec.webSearchModel &&
+			imageModel === spec.imageModel
+		) {
+			continue;
+		}
+		specs[index] = { ...spec, contextPromotionTarget, compactionModel, webSearchModel, imageModel };
 	}
 }
 

package/src/compat/resolve.ts

@@ -31,7 +31,12 @@ import type {
 	ResolvedOpenAISharedCompat,
 	ThinkingConfig,
 } from "../types";
-import { isAnthropicSigningProxyUrl, isAzureAnthropicRoute, isOfficialAnthropicApiUrl } from "./anthropic";
+import {
+	isAnthropicSigningProxyUrl,
+	isAzureAnthropicRoute,
+	isBedrockAnthropicRoute,
+	isOfficialAnthropicApiUrl,
+} from "./anthropic";
 import { applyCompatOverrides } from "./apply";
 import { API_COMPAT_RECORDS, AXES, type CompatRecordName } from "./axes";
 import { hasModelScopedEffortsRule, resolveCascade } from "./cascade";
@@ -853,7 +858,9 @@ function resolveAnthropicPolicy(
 	const isCopilot = modelMatchesHost(spec, "githubCopilot");
 	const isZenmux = modelMatchesHost(spec, "zenmux");
 	const requiresThinkingEnabled = modelMatchesHost(spec, "moonshotNative") && facts.kimiMandatoryThinking;
-	const isAzure = isAzureAnthropicRoute(baseUrl);
+	const bedrockMessagesApi = isBedrockAnthropicRoute(baseUrl);
+	// Both routes reject the top-level tool `strict` field.
+	const rejectsStrictTools = isAzureAnthropicRoute(baseUrl) || bedrockMessagesApi;
 	const signingEndpoint = official || isCopilot || isZenmux || isAnthropicSigningProxyUrl(baseUrl);
 	const compat: ResolvedAnthropicCompat = {
 		officialEndpoint: official,
@@ -862,7 +869,9 @@ function resolveAnthropicPolicy(
 		supportsServerCompaction: false,
 		firstPartyProvider: false,
 		supportsOutputEffort: true,
-		disableStrictTools: isAzure,
+		disableStrictTools: rejectsStrictTools,
+		// Present as a key so models.yml `compat` can set it; unset unless detected.
+		bedrockMessagesApi: bedrockMessagesApi || undefined,
 		disableAdaptiveThinking: false,
 		allowAnthropicHeaderOverrides: false,
 		supportsEagerToolInputStreaming: official,
@@ -872,6 +881,7 @@ function resolveAnthropicPolicy(
 		supportsMidConversationToolChanges: false,
 		supportsPerMessageEffort: false,
 		supportsThinkingBindingControls: false,
+		supportsBetweenToolsThinking: false,
 		supportsForcedToolChoice: !requiresThinkingEnabled && !facts.family("fable", "mythos"),
 		supportsSamplingParams: !facts.anthropicAdaptiveGenAtLeast("4.7"),
 		requiresToolResultId: false,

package/src/compat/rules/README.md

@@ -358,7 +358,7 @@ auth "anthropic" {
 
 Login kinds:
 
-- `login "api-key" { auth-url "…"; instructions "…"; prompt "…" placeholder="…"; empty-fallback "…"; normalize "strip-bearer"; validate … }` — `validate "chat-completions" base-url= model= tolerate-model-denied= max-tokens-field= max-tokens=`, `validate "anthropic-messages" base-url= model=`, or `validate "models-endpoint" url= base-url-env= headers-hook=`; all accept `label=` (error-message label, defaults to `name`) and `optional=#true` (only auth failures reject).
+- `login "api-key" { auth-url "…"; instructions "…"; prompt "…" placeholder="…"; empty-fallback "…"; normalize "strip-bearer"; validate … }` — `validate "chat-completions" base-url= model= tolerate-model-denied= max-tokens-field= max-tokens=`, `validate "anthropic-messages" base-url= model=`, or `validate "models-endpoint" url= base-url-env= headers-hook=`; all accept `label=` (error-message label, defaults to `name`), `optional=#true` (only auth failures reject), and `trust-forbidden=#true` (requires `optional=#true`; a 403 no longer rejects, so only a 401 does).
 - `login "oauth-code" { … }` as above; `token`/`refresh` `params` values may use `{code}`, `{state}`, `{redirect_uri}`, `{code_verifier}`, `{client_id}`, `{client_secret}`, `{refresh_token}`, `{scope}`; the standard grant parameters are sent unless `standard=#false`.
 - `login "device-code" { client-id …; base-url "…"; scopes …; headers-hook "…"; device url="{base}/…" body="form" { params {…} headers {…} }; token url="…" url-hook="…"; response user-code= device-code= verification-uri= verification-uri-complete= interval= expires-in=; instructions "Enter code: {user_code}"; credential {…}; userinfo …; after-exchange hook=… }`.
 - `login "custom" hook="name"` — the whole flow is a named `@oh-my-pi/pi-ai` hook (`src/registry/hooks/custom.ts`).

package/src/compat/rules/auth/commandcode.kdl

@@ -1,7 +1,12 @@
-// No `validate` probe on purpose: GET /provider/v1/models is public (200
-// with no or bogus credentials, verified 2026-09-09), so a models-endpoint
-// probe could never reject a bad key. A chat-completions probe would bill
-// the key and 403 for Go-plan keys that are still valid CLI credentials.
+// `/provider/v1/models` is public: it answers 200 to a missing or bogus key.
+// Login therefore validates against `GET /alpha/whoami`, the identity route
+// that command-code@1.67.0 calls. That route answers 401 to a bad key
+// (verified 2026-09-29) and bills nothing. It is not in the Provider API
+// docs, so the probe is optional and only a 401 rejects the key, matching
+// `validateCommandApiKey` in command-code@1.67.0. A 403, a 404, a 5xx, or a
+// moved route trusts the pasted key.
+// The `models-endpoint` probe only checks for a 2xx answer, so this identity
+// route serves as its target.
 auth "commandcode" {
 	name "Command Code"
 	env "COMMAND_CODE_API_KEY" "COMMANDCODE_API_KEY"
@@ -9,5 +14,6 @@ auth "commandcode" {
 		auth-url "https://commandcode.ai/studio"
 		instructions "Create or copy a Provider API key from Command Code Studio"
 		prompt "Paste your Command Code API key" placeholder="user_..."
+		validate "models-endpoint" label="Command Code" url="https://api.commandcode.ai/alpha/whoami" optional=#true trust-forbidden=#true
 	}
 }

package/src/compat/rules/classes/anthropic.kdl

@@ -7,11 +7,11 @@ class "anthropic" {
 			thinking-prefix-binding #true
 		}
 	}
-	// On-demand compaction (compact-2026-09-04) is available on the Claude
-	// API and Vertex, not Amazon Bedrock or unverified gateway deployments.
+	// On-demand compaction (compact-2026-09-04) works on the Claude API, Vertex,
+	// and Amazon Bedrock's `/anthropic` routes, not Bedrock Converse.
 	// Foundry / Claude Platform on AWS are first-party Anthropic routes whose
 	// effective URLs are checked at request time.
-	on "anthropic" "google-vertex" {
+	on "anthropic" "google-vertex" "amazon-bedrock" "bedrock-mantle" {
 		family "fable" {
 			revision ">=5" {
 				supports-server-compaction #true
@@ -61,6 +61,21 @@ class "anthropic" {
 			supports-forced-tool-choice #false
 		}
 	}
+	// Sonnet 5.5 rejects forced tool selection like Opus 5.5 (400 `tool_choice:
+	// type "tool" and "any" are not supported for this model`); upper-bounded
+	// for the same reason as the Opus rule above.
+	// https://platform.claude.com/docs/en/models/sonnet-5-5/whats-new-sonnet-5-5
+	family "sonnet" {
+		revision ">=5.5 <6" {
+			supports-forced-tool-choice #false
+			// `thinking: {type: "disabled"}` is a 400; `between_tools` is the
+			// lowest thinking setting.
+			// https://platform.claude.com/docs/en/models/sonnet-5-5/migration-guide#turn-off-up-front-thinking
+			supports-between-tools-thinking #true
+			// Signed thinking is bound to the exact preceding conversation.
+			thinking-prefix-binding #true
+		}
+	}
 	on "anthropic" "cloudflare-ai-gateway" "google-vertex" {
 		family "fable" {
 			revision ">=5" {
@@ -87,6 +102,15 @@ class "anthropic" {
 				supports-thinking-binding-controls #true
 			}
 		}
+		family "sonnet" {
+			revision ">=5.5" {
+				supports-mid-conversation-system #true
+				supports-mid-conversation-tool-changes #true
+				supports-turn-scoped-system #true
+				supports-per-message-effort #true
+				supports-thinking-binding-controls #true
+			}
+		}
 		family "opus" {
 			revision ">=4.8" {
 				supports-mid-conversation-system #true
@@ -369,3 +393,13 @@ class "anthropic" {
 		antigravity-usage-label "true"
 	}
 }
+
+// Claude 4+ ships the server-side `web_search` tool on any Messages-compatible host.
+class "anthropic" {
+	on-api "anthropic-messages" {
+		revision ">=4" {
+			web-search "anthropic"
+			web-search-model "claude-haiku-4-5"
+		}
+	}
+}

package/src/compat/rules/classes/gemini.kdl

@@ -107,3 +107,14 @@ class "gemini" {
 
 	thinking-loop-guard "gemini"
 }
+
+// Google Search grounding is a Gemini 2+ lineage capability on every Gemini transport.
+class "gemini" {
+	on-api "google-generative-ai" "google-gemini-cli" {
+		revision ">=2" {
+			web-search "gemini"
+			web-search-model "gemini-2.5-flash"
+			image-model "gemini-3-pro-image"
+		}
+	}
+}

package/src/compat/rules/classes/openai.kdl

@@ -59,3 +59,25 @@ class "openai" {
 		}
 	}
 }
+
+// Hosted Responses tools are a GPT-5+ lineage capability on any
+// Responses-compatible host, so proxies exposing these ids inherit them. The
+// swap targets resolve on the same provider only; when a host lacks them the
+// session model runs the tool itself.
+class "openai" {
+	on-api "openai-responses" {
+		revision ">=5" {
+			web-search "openai"
+			web-search-model "gpt-5.6-luna"
+			hosted-image #true
+			image-model "gpt-image-2"
+		}
+	}
+	on-api "openai-codex-responses" {
+		revision ">=5" {
+			web-search-model "gpt-5.6-luna"
+			hosted-image #true
+			image-model "gpt-image-2"
+		}
+	}
+}

package/src/compat/rules/classes/xai.kdl

@@ -43,3 +43,9 @@ class "xai" {
 	}
 	thinking-loop-guard "xai"
 }
+
+// Same-provider swap targets for hosted search and image generation.
+class "xai" {
+	web-search-model "grok-4.5"
+	image-model "grok-imagine-image"
+}

package/src/compat/rules/providers/anthropic.kdl

@@ -28,6 +28,12 @@ provider "anthropic" {
 			cost input=3 output=15 cache-read=0.3 cache-write=3.75
 			limits context=1000000 max-tokens=128000
 		}
+		model "claude-sonnet-5-5" name="Claude Sonnet 5.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=2 output=10 cache-read=0.2 cache-write=2.5
+			limits context=1000000 max-tokens=128000
+		}
 		model "claude-fable-5" name="Claude Fable 5" {
 			reasoning #true
 			input "text" "image"