gsd-pi 2.75.0 → 2.76.0

package/packages/pi-ai/src/models/generated/groq.ts

@@ -4,40 +4,6 @@
 import type { Model } from "../../types.js";
 
 export const GROQ_MODELS = {
-		"deepseek-r1-distill-llama-70b": {
-			id: "deepseek-r1-distill-llama-70b",
-			name: "DeepSeek R1 Distill Llama 70B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: true,
-			input: ["text"],
-			cost: {
-				input: 0.75,
-				output: 0.99,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 131072,
-			maxTokens: 8192,
-		} satisfies Model<"openai-completions">,
-		"gemma2-9b-it": {
-			id: "gemma2-9b-it",
-			name: "Gemma 2 9B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 0.2,
-				output: 0.2,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 8192,
-			maxTokens: 8192,
-		} satisfies Model<"openai-completions">,
 		"groq/compound": {
 			id: "groq/compound",
 			name: "Compound",
@@ -106,57 +72,6 @@ export const GROQ_MODELS = {
 			contextWindow: 131072,
 			maxTokens: 32768,
 		} satisfies Model<"openai-completions">,
-		"llama3-70b-8192": {
-			id: "llama3-70b-8192",
-			name: "Llama 3 70B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 0.59,
-				output: 0.79,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 8192,
-			maxTokens: 8192,
-		} satisfies Model<"openai-completions">,
-		"llama3-8b-8192": {
-			id: "llama3-8b-8192",
-			name: "Llama 3 8B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 0.05,
-				output: 0.08,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 8192,
-			maxTokens: 8192,
-		} satisfies Model<"openai-completions">,
-		"meta-llama/llama-4-maverick-17b-128e-instruct": {
-			id: "meta-llama/llama-4-maverick-17b-128e-instruct",
-			name: "Llama 4 Maverick 17B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text", "image"],
-			cost: {
-				input: 0.2,
-				output: 0.6,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 131072,
-			maxTokens: 8192,
-		} satisfies Model<"openai-completions">,
 		"meta-llama/llama-4-scout-17b-16e-instruct": {
 			id: "meta-llama/llama-4-scout-17b-16e-instruct",
 			name: "Llama 4 Scout 17B",
@@ -174,57 +89,6 @@ export const GROQ_MODELS = {
 			contextWindow: 131072,
 			maxTokens: 8192,
 		} satisfies Model<"openai-completions">,
-		"mistral-saba-24b": {
-			id: "mistral-saba-24b",
-			name: "Mistral Saba 24B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 0.79,
-				output: 0.79,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 32768,
-			maxTokens: 32768,
-		} satisfies Model<"openai-completions">,
-		"moonshotai/kimi-k2-instruct": {
-			id: "moonshotai/kimi-k2-instruct",
-			name: "Kimi K2 Instruct",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 1,
-				output: 3,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 131072,
-			maxTokens: 16384,
-		} satisfies Model<"openai-completions">,
-		"moonshotai/kimi-k2-instruct-0905": {
-			id: "moonshotai/kimi-k2-instruct-0905",
-			name: "Kimi K2 Instruct 0905",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: false,
-			input: ["text"],
-			cost: {
-				input: 1,
-				output: 3,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 262144,
-			maxTokens: 16384,
-		} satisfies Model<"openai-completions">,
 		"openai/gpt-oss-120b": {
 			id: "openai/gpt-oss-120b",
 			name: "GPT OSS 120B",
@@ -276,23 +140,6 @@ export const GROQ_MODELS = {
 			contextWindow: 131072,
 			maxTokens: 65536,
 		} satisfies Model<"openai-completions">,
-		"qwen-qwq-32b": {
-			id: "qwen-qwq-32b",
-			name: "Qwen QwQ 32B",
-			api: "openai-completions",
-			provider: "groq",
-			baseUrl: "https://api.groq.com/openai/v1",
-			reasoning: true,
-			input: ["text"],
-			cost: {
-				input: 0.29,
-				output: 0.39,
-				cacheRead: 0,
-				cacheWrite: 0,
-			},
-			contextWindow: 131072,
-			maxTokens: 16384,
-		} satisfies Model<"openai-completions">,
 		"qwen/qwen3-32b": {
 			id: "qwen/qwen3-32b",
 			name: "Qwen3 32B",

package/packages/pi-ai/src/models/generated/openai-codex.ts

@@ -140,4 +140,21 @@ export const OPENAI_CODEX_MODELS = {
 			contextWindow: 272000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-codex-responses">,
+		"gpt-5.4-mini": {
+			id: "gpt-5.4-mini",
+			name: "GPT-5.4 Mini",
+			api: "openai-codex-responses",
+			provider: "openai-codex",
+			baseUrl: "https://chatgpt.com/backend-api",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 0.75,
+				output: 4.5,
+				cacheRead: 0.075,
+				cacheWrite: 0,
+			},
+			contextWindow: 272000,
+			maxTokens: 128000,
+		} satisfies Model<"openai-codex-responses">,
 	} as const satisfies Record<string, Model<any>>;

package/packages/pi-ai/src/models/generated/openrouter.ts

@@ -293,6 +293,23 @@ export const OPENROUTER_MODELS = {
 			contextWindow: 1000000,
 			maxTokens: 128000,
 		} satisfies Model<"openai-completions">,
+		"anthropic/claude-opus-4.7": {
+			id: "anthropic/claude-opus-4.7",
+			name: "Anthropic: Claude Opus 4.7",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			reasoning: true,
+			input: ["text", "image"],
+			cost: {
+				input: 5,
+				output: 25,
+				cacheRead: 0.5,
+				cacheWrite: 6.25,
+			},
+			contextWindow: 1000000,
+			maxTokens: 128000,
+		} satisfies Model<"openai-completions">,
 		"anthropic/claude-opus-4.6-fast": {
 			id: "anthropic/claude-opus-4.6-fast",
 			name: "Anthropic: Claude Opus 4.6 (Fast)",

package/packages/pi-ai/src/models.generated.test.ts

@@ -326,6 +326,16 @@ describe("removed models are absent from the registry", () => {
 		{ provider: "openrouter", id: "meituan/longcat-flash-thinking" },
 		{ provider: "vercel-ai-gateway", id: "xai/grok-2-vision" },
 		{ provider: "anthropic", id: "claude-3-7-sonnet-latest" },
+		// Groq decommissioned models — issue #4257
+		{ provider: "groq", id: "llama3-70b-8192" },
+		{ provider: "groq", id: "llama3-8b-8192" },
+		{ provider: "groq", id: "deepseek-r1-distill-llama-70b" },
+		{ provider: "groq", id: "gemma2-9b-it" },
+		{ provider: "groq", id: "meta-llama/llama-4-maverick-17b-128e-instruct" },
+		{ provider: "groq", id: "mistral-saba-24b" },
+		{ provider: "groq", id: "moonshotai/kimi-k2-instruct" },
+		{ provider: "groq", id: "moonshotai/kimi-k2-instruct-0905" },
+		{ provider: "groq", id: "qwen-qwq-32b" },
 	];
 
 	for (const { provider, id } of removedModels) {
@@ -349,11 +359,18 @@ describe("spot-checks for models added in this regeneration", () => {
 		{ provider: "openrouter", id: "arcee-ai/trinity-large-thinking", reasoning: true },
 		{ provider: "openrouter", id: "openai/gpt-audio" },
 		{ provider: "openrouter", id: "anthropic/claude-opus-4.6-fast" },
+		{ provider: "openrouter", id: "anthropic/claude-opus-4.7" },
+		{ provider: "amazon-bedrock", id: "anthropic.claude-opus-4-7" },
+		{ provider: "amazon-bedrock", id: "us.anthropic.claude-opus-4-7" },
+		{ provider: "amazon-bedrock", id: "eu.anthropic.claude-opus-4-7" },
+		{ provider: "amazon-bedrock", id: "global.anthropic.claude-opus-4-7" },
+		{ provider: "google-antigravity", id: "claude-opus-4-7-thinking", reasoning: true },
 		{ provider: "openrouter", id: "qwen/qwen3.6-plus" },
 		{ provider: "groq", id: "groq/compound" },
 		{ provider: "groq", id: "groq/compound-mini" },
 		{ provider: "huggingface", id: "zai-org/GLM-5.1" },
 		{ provider: "openai", id: "gpt-5.3-chat-latest" },
+		{ provider: "openai-codex", id: "gpt-5.4-mini", reasoning: true },
 		{ provider: "mistral", id: "mistral-small-2603" },
 		{ provider: "zai", id: "glm-5.1" },
 	];

package/packages/pi-ai/src/providers/amazon-bedrock.test.ts

@@ -0,0 +1,164 @@
+/**
+ * Regression guards for #4392
+ *
+ * Verifies that `supportsAdaptiveThinking()` in amazon-bedrock.ts correctly
+ * recognises all current adaptive-thinking-capable models: opus-4-7,
+ * sonnet-4-7, and haiku-4-5 (in addition to the previously supported
+ * opus-4-6 / sonnet-4-6 family).
+ *
+ * Related: #4392 (opus-4-7 adaptive thinking not recognised on Bedrock)
+ *          #4352 (pre-existing: only opus-4-6 / sonnet-4-6 whitelisted)
+ */
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+
+import {
+    supportsAdaptiveThinking,
+    mapThinkingLevelToEffort,
+    buildAdditionalModelRequestFields,
+    type BedrockOptions,
+} from "./amazon-bedrock.js";
+
+import type { Model } from "../types.js";
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+/** Create a minimal stub `Model` for the given ID to use in unit tests. */
+function makeModel(id: string): Model<"bedrock-converse-stream"> {
+    return {
+        id,
+        name: id,
+        api: "bedrock-converse-stream",
+        provider: "amazon-bedrock" as any,
+        baseUrl: "",
+        reasoning: true,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: 200000,
+        maxTokens: 32000,
+    };
+}
+
+// ---------------------------------------------------------------------------
+// supportsAdaptiveThinking — regression guards (#4392 / #4352)
+// ---------------------------------------------------------------------------
+
+describe("supportsAdaptiveThinking — regression guard #4392 / #4352: all supported models", () => {
+    it("returns true for opus-4-6 (hyphen, Bedrock ARN style)", () => {
+        assert.ok(supportsAdaptiveThinking("anthropic.claude-opus-4-6-20250514-v1:0"));
+    });
+
+    it("returns true for sonnet-4-6 (hyphen)", () => {
+        assert.ok(supportsAdaptiveThinking("anthropic.claude-sonnet-4-6-20250514-v1:0"));
+    });
+
+    it("[#4392] returns true for opus-4-7 (hyphen, Bedrock ARN style)", () => {
+        assert.ok(
+            supportsAdaptiveThinking("anthropic.claude-opus-4-7-20250514-v1:0"),
+            "opus-4-7 should support adaptive thinking (regression guard #4392)",
+        );
+    });
+
+    it("[#4392] returns true for opus-4-7 (dot separator)", () => {
+        assert.ok(
+            supportsAdaptiveThinking("anthropic.claude-opus-4.7-20250514-v1:0"),
+            "opus-4.7 (dot) should support adaptive thinking (regression guard #4392)",
+        );
+    });
+
+    it("[#4352] returns true for sonnet-4-7 (hyphen)", () => {
+        assert.ok(
+            supportsAdaptiveThinking("anthropic.claude-sonnet-4-7-20250514-v1:0"),
+            "sonnet-4-7 should support adaptive thinking (regression guard #4352)",
+        );
+    });
+
+    it("[#4352] returns true for haiku-4-5 (hyphen)", () => {
+        assert.ok(
+            supportsAdaptiveThinking("anthropic.claude-haiku-4-5-20250514-v1:0"),
+            "haiku-4-5 should support adaptive thinking (regression guard #4352)",
+        );
+    });
+});
+
+// ---------------------------------------------------------------------------
+// buildAdditionalModelRequestFields — adaptive thinking output for opus-4-7
+// Tests go through the public API surface to validate end-to-end behaviour.
+// ---------------------------------------------------------------------------
+
+describe("buildAdditionalModelRequestFields — Bug #4392: opus-4-7 must use adaptive thinking", () => {
+    const options: BedrockOptions = { reasoning: "high" };
+
+    it("[#4392] opus-4-7 Bedrock ARN → thinking.type === 'adaptive' (not budget_tokens)", () => {
+        const model = makeModel("anthropic.claude-opus-4-7-20250514-v1:0");
+        const fields = buildAdditionalModelRequestFields(model, options);
+        // Regression guard: supportsAdaptiveThinking must return true for opus-4-7
+        // so the function produces thinking.type='adaptive', not budget_tokens.
+        assert.equal(
+            fields?.thinking?.type,
+            "adaptive",
+            "opus-4-7 should produce thinking.type='adaptive', not budget_tokens",
+        );
+    });
+
+    it("[#4392] opus-4-7 dot separator → thinking.type === 'adaptive'", () => {
+        const model = makeModel("anthropic.claude-opus-4.7-20250514-v1:0");
+        const fields = buildAdditionalModelRequestFields(model, options);
+        assert.equal(
+            fields?.thinking?.type,
+            "adaptive",
+            "opus-4.7 (dot) should produce thinking.type='adaptive'",
+        );
+    });
+
+    it("[#4352] sonnet-4-7 → thinking.type === 'adaptive'", () => {
+        const model = makeModel("anthropic.claude-sonnet-4-7-20250514-v1:0");
+        const fields = buildAdditionalModelRequestFields(model, options);
+        assert.equal(
+            fields?.thinking?.type,
+            "adaptive",
+            "sonnet-4-7 should produce thinking.type='adaptive'",
+        );
+    });
+
+    it("[#4352] haiku-4-5 → thinking.type === 'adaptive'", () => {
+        const model = makeModel("anthropic.claude-haiku-4-5-20250514-v1:0");
+        const fields = buildAdditionalModelRequestFields(model, options);
+        assert.equal(
+            fields?.thinking?.type,
+            "adaptive",
+            "haiku-4-5 should produce thinking.type='adaptive'",
+        );
+    });
+});
+
+// ---------------------------------------------------------------------------
+// mapThinkingLevelToEffort — regression guard for xhigh on opus-4-7 (#4392)
+// Fixed: Bedrock provider now returns "xhigh" for 4.7+ models instead of "max".
+// ---------------------------------------------------------------------------
+
+describe("mapThinkingLevelToEffort — regression guard #4392: opus-4-7 xhigh returns 'xhigh'", () => {
+    it("[#4392] maps xhigh → 'xhigh' for opus-4-7 (native xhigh support)", () => {
+        // Regression guard: mapThinkingLevelToEffort must return "xhigh" for opus-4-7,
+        // not "max". Ensures the fix in #4392 does not regress.
+        const result = mapThinkingLevelToEffort("xhigh", "anthropic.claude-opus-4-7-20250514-v1:0");
+        assert.equal(
+            result,
+            "xhigh",
+            "opus-4-7 supports native xhigh effort — must not be clamped to 'max'",
+        );
+    });
+
+    it("[#4392] maps xhigh → 'max' for opus-4-6 (no native xhigh, clamped)", () => {
+        // This already passes — regression guard.
+        const result = mapThinkingLevelToEffort("xhigh", "anthropic.claude-opus-4-6-20250514-v1:0");
+        assert.equal(result, "max");
+    });
+
+    it("maps high → 'high' for opus-4-7 (not affected by bug)", () => {
+        const result = mapThinkingLevelToEffort("high", "anthropic.claude-opus-4-7-20250514-v1:0");
+        assert.equal(result, "high");
+    });
+});

package/packages/pi-ai/src/providers/amazon-bedrock.ts

@@ -45,6 +45,7 @@ import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
 import { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from "./simple-options.js";
 import { transformMessagesWithReport } from "./transform-messages.js";
 
+/** Stream options specific to the Amazon Bedrock converse-stream provider, including region, reasoning, and caching knobs. */
 export interface BedrockOptions extends StreamOptions {
 	region?: string;
 	profile?: string;
@@ -57,8 +58,10 @@ export interface BedrockOptions extends StreamOptions {
 	interleavedThinking?: boolean;
 }
 
+/** Internal working type that annotates content blocks with a streaming index and partial JSON accumulator. */
 type Block = (TextContent | ThinkingContent | ToolCall) & { index?: number; partialJson?: string };
 
+/** Stream a conversation turn via Amazon Bedrock's converse-stream API. */
 export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOptions> = (
 	model: Model<"bedrock-converse-stream">,
 	context: Context,
@@ -216,6 +219,7 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOpt
 	return stream;
 };
 
+/** Simplified entry point for Bedrock streaming; resolves thinking budgets and adaptive-thinking support. */
 export const streamSimpleBedrock: StreamFunction<"bedrock-converse-stream", SimpleStreamOptions> = (
 	model: Model<"bedrock-converse-stream">,
 	context: Context,
@@ -260,6 +264,7 @@ export const streamSimpleBedrock: StreamFunction<"bedrock-converse-stream", Simp
 	} satisfies BedrockOptions);
 };
 
+/** Handle a `contentBlockStart` event, initialising a new tool-call block when a tool-use start arrives. */
 function handleContentBlockStart(
 	event: ContentBlockStartEvent,
 	blocks: Block[],
@@ -283,6 +288,7 @@ function handleContentBlockStart(
 	}
 }
 
+/** Handle a `contentBlockDelta` event, appending text, tool-input JSON, or reasoning content to the active block. */
 function handleContentBlockDelta(
 	event: ContentBlockDeltaEvent,
 	blocks: Block[],
@@ -341,6 +347,7 @@ function handleContentBlockDelta(
 	}
 }
 
+/** Handle a `metadata` event, updating token-usage counters and cost on the output message. */
 function handleMetadata(
 	event: ConverseStreamMetadataEvent,
 	model: Model<"bedrock-converse-stream">,
@@ -356,6 +363,7 @@ function handleMetadata(
 	}
 }
 
+/** Handle a `contentBlockStop` event, finalising the block and pushing the appropriate completion event. */
 function handleContentBlockStop(
 	event: ContentBlockStopEvent,
 	blocks: Block[],
@@ -383,21 +391,33 @@ function handleContentBlockStop(
 }
 
 /**
- * Check if the model supports adaptive thinking (Opus 4.6 and Sonnet 4.6).
+ * Check if the model supports adaptive thinking (Opus 4.6/4.7, Sonnet 4.6/4.7, Haiku 4.5).
+ * @internal exported for testing only
  */
-function supportsAdaptiveThinking(modelId: string): boolean {
+export function supportsAdaptiveThinking(modelId: string): boolean {
 	return (
 		modelId.includes("opus-4-6") ||
 		modelId.includes("opus-4.6") ||
+		modelId.includes("opus-4-7") ||
+		modelId.includes("opus-4.7") ||
 		modelId.includes("sonnet-4-6") ||
-		modelId.includes("sonnet-4.6")
+		modelId.includes("sonnet-4.6") ||
+		modelId.includes("sonnet-4-7") ||
+		modelId.includes("sonnet-4.7") ||
+		modelId.includes("haiku-4-5") ||
+		modelId.includes("haiku-4.5")
 	);
 }
 
-function mapThinkingLevelToEffort(
+/**
+ * Maps a reasoning/thinking level to the Bedrock effort string for the given model.
+ * Returns `"xhigh"` for 4.7+ models and `"max"` for older ones; `"low"` for minimal/low.
+ * @internal exported for testing only
+ */
+export function mapThinkingLevelToEffort(
 	level: SimpleStreamOptions["reasoning"],
 	modelId: string,
-): "low" | "medium" | "high" | "max" {
+): "low" | "medium" | "high" | "xhigh" | "max" {
 	switch (level) {
 		case "minimal":
 		case "low":
@@ -407,7 +427,9 @@ function mapThinkingLevelToEffort(
 		case "high":
 			return "high";
 		case "xhigh":
-			return modelId.includes("opus-4-6") || modelId.includes("opus-4.6") ? "max" : "high";
+			if (modelId.includes("opus-4-7") || modelId.includes("opus-4.7")) return "xhigh";
+			if (modelId.includes("opus-4-6") || modelId.includes("opus-4.6")) return "max";
+			return "high";
 		default:
 			return "high";
 	}
@@ -457,6 +479,7 @@ function supportsThinkingSignature(model: Model<"bedrock-converse-stream">): boo
 	return id.includes("anthropic.claude") || id.includes("anthropic/claude");
 }
 
+/** Build the Bedrock system-prompt block array, appending a cache point for supported models when caching is enabled. */
 function buildSystemPrompt(
 	systemPrompt: string | undefined,
 	model: Model<"bedrock-converse-stream">,
@@ -476,11 +499,13 @@ function buildSystemPrompt(
 	return blocks;
 }
 
+/** Sanitise a tool-call ID to alphanumeric, underscore, and hyphen characters (max 64 chars) for Bedrock compatibility. */
 function normalizeToolCallId(id: string): string {
 	const sanitized = id.replace(/[^a-zA-Z0-9_-]/g, "_");
 	return sanitized.length > 64 ? sanitized.slice(0, 64) : sanitized;
 }
 
+/** Convert GSD context messages to the Bedrock `Message[]` format, collapsing consecutive tool-result turns into a single user message. */
 function convertMessages(
 	context: Context,
 	model: Model<"bedrock-converse-stream">,
@@ -630,6 +655,7 @@ function convertMessages(
 	return result;
 }
 
+/** Convert GSD tool definitions and tool-choice preference to a Bedrock `ToolConfiguration`, appending a cache point for supported models. */
 function convertToolConfig(
 	tools: Tool[] | undefined,
 	toolChoice: BedrockOptions["toolChoice"],
@@ -673,6 +699,7 @@ function convertToolConfig(
 	return { tools: bedrockTools, toolChoice: bedrockToolChoice };
 }
 
+/** Map a Bedrock stop-reason string to GSD's internal `StopReason`. */
 function mapStopReason(reason: string | undefined): StopReason {
 	switch (reason) {
 		case BedrockStopReason.END_TURN:
@@ -688,7 +715,13 @@ function mapStopReason(reason: string | undefined): StopReason {
 	}
 }
 
-function buildAdditionalModelRequestFields(
+/**
+ * Builds the Bedrock `additionalModelRequestFields` payload for Claude models.
+ * Handles adaptive vs. budget-based thinking, beta flags, and xhigh-to-max clamping
+ * for models that lack native xhigh support.
+ * @internal exported for testing only
+ */
+export function buildAdditionalModelRequestFields(
 	model: Model<"bedrock-converse-stream">,
 	options: BedrockOptions,
 ): Record<string, any> | undefined {
@@ -733,6 +766,7 @@ function buildAdditionalModelRequestFields(
 	return undefined;
 }
 
+/** Convert a base64-encoded image to a Bedrock image content block with the appropriate `ImageFormat`. */
 function createImageBlock(mimeType: string, data: string) {
 	let format: ImageFormat;
 	switch (mimeType) {

package/packages/pi-ai/src/providers/anthropic-auth.test.ts

@@ -4,7 +4,7 @@ import { readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
 
-import { usesAnthropicBearerAuth } from "./anthropic.js";
+import { usesAnthropicBearerAuth, resolveAnthropicBaseUrl } from "./anthropic.js";
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 
@@ -30,3 +30,49 @@ test("createClient routes Bearer-auth providers through authToken (#3783)", () =
 		"Bearer-auth providers should send authToken instead",
 	);
 });
+
+// Minimal model stub — only the field resolveAnthropicBaseUrl cares about.
+const stubModel = { baseUrl: "https://api.anthropic.com" } as Parameters<typeof resolveAnthropicBaseUrl>[0];
+
+test("resolveAnthropicBaseUrl returns model.baseUrl when ANTHROPIC_BASE_URL is unset (#4140)", (t) => {
+	const saved = process.env.ANTHROPIC_BASE_URL;
+	t.after(() => {
+		if (saved === undefined) delete process.env.ANTHROPIC_BASE_URL;
+		else process.env.ANTHROPIC_BASE_URL = saved;
+	});
+
+	delete process.env.ANTHROPIC_BASE_URL;
+	assert.equal(resolveAnthropicBaseUrl(stubModel), "https://api.anthropic.com");
+});
+
+test("resolveAnthropicBaseUrl prefers ANTHROPIC_BASE_URL over model.baseUrl (#4140)", (t) => {
+	const saved = process.env.ANTHROPIC_BASE_URL;
+	t.after(() => {
+		if (saved === undefined) delete process.env.ANTHROPIC_BASE_URL;
+		else process.env.ANTHROPIC_BASE_URL = saved;
+	});
+
+	process.env.ANTHROPIC_BASE_URL = "https://proxy.example.com";
+	assert.equal(resolveAnthropicBaseUrl(stubModel), "https://proxy.example.com");
+});
+
+test("resolveAnthropicBaseUrl ignores whitespace-only ANTHROPIC_BASE_URL (#4140)", (t) => {
+	const saved = process.env.ANTHROPIC_BASE_URL;
+	t.after(() => {
+		if (saved === undefined) delete process.env.ANTHROPIC_BASE_URL;
+		else process.env.ANTHROPIC_BASE_URL = saved;
+	});
+
+	process.env.ANTHROPIC_BASE_URL = "   ";
+	assert.equal(resolveAnthropicBaseUrl(stubModel), "https://api.anthropic.com");
+});
+
+test("createClient uses resolveAnthropicBaseUrl for all auth paths (#4140)", () => {
+	const source = readFileSync(join(__dirname, "..", "..", "src", "providers", "anthropic.ts"), "utf-8");
+	const directUsages = (source.match(/baseURL:\s*model\.baseUrl/g) ?? []).length;
+	assert.equal(directUsages, 0, "createClient must not use model.baseUrl directly — use resolveAnthropicBaseUrl(model)");
+	assert.ok(
+		source.includes("baseURL: resolveAnthropicBaseUrl(model)"),
+		"all createClient branches should pass baseURL through resolveAnthropicBaseUrl",
+	);
+});

package/packages/pi-ai/src/providers/anthropic-shared.test.ts

@@ -1,6 +1,6 @@
 import { describe, it } from "node:test";
 import assert from "node:assert/strict";
-import { convertTools, mapStopReason } from "./anthropic-shared.js";
+import { convertTools, mapStopReason, mapThinkingLevelToEffort } from "./anthropic-shared.js";
 
 const makeTool = (name: string) =>
 	({
@@ -56,6 +56,20 @@ describe("convertTools cache_control", () => {
 	});
 });
 
+describe("mapThinkingLevelToEffort", () => {
+	it("maps xhigh to max for opus-4-6 (no native xhigh support)", () => {
+		assert.equal(mapThinkingLevelToEffort("xhigh", "claude-opus-4-6"), "max");
+	});
+
+	it("maps xhigh to xhigh natively for opus-4-7", () => {
+		assert.equal(mapThinkingLevelToEffort("xhigh", "claude-opus-4-7"), "xhigh");
+	});
+
+	it("maps high to high for opus-4-7", () => {
+		assert.equal(mapThinkingLevelToEffort("high", "claude-opus-4-7"), "high");
+	});
+});
+
 describe("mapStopReason", () => {
 	it("maps end_turn to stop", () => {
 		assert.equal(mapStopReason("end_turn"), "stop");