@pensar/apex 2.6.1 → 2.7.0-canary.1572d9a4

package/src/core/ai/compactionTelemetry.test.ts

@@ -45,7 +45,7 @@ function model() {
       stream: simulateReadableStream({
         chunks: [
           { type: "text-start", id: "t" },
-          { type: "text-delta", id: "t", delta: "Done" },
+          { type: "text-delta", id: "t", delta: "A compact summary" },
           { type: "text-end", id: "t" },
           {
             type: "finish",
@@ -58,6 +58,26 @@ function model() {
   });
 }
 
+const SUMMARIZE_FN = "apex.context.summarize";
+
+// Summarization and the resumed agent turn both reach the model through doStream,
+// so a mock that must fail only the summary matches on its instruction.
+function isSummaryCall(options: { prompt: unknown }): boolean {
+  return JSON.stringify(options.prompt).includes("Compact this conversation");
+}
+
+// Located by telemetry functionId rather than span name, so it survives the
+// summarizer switching between streaming and one-shot generation.
+function summarySpan<
+  T extends { name: string; attributes: Record<string, unknown> },
+>(spans: readonly T[]): T | undefined {
+  return spans.find(
+    (span) =>
+      span.name.endsWith(".doStream") &&
+      span.attributes["ai.telemetry.functionId"] === SUMMARIZE_FN,
+  );
+}
+
 async function drain(stream: { fullStream: AsyncIterable<unknown> }) {
   for await (const _part of stream.fullStream) {
     /* consume the SDK lifecycle */
@@ -193,10 +213,13 @@ describe("compaction observability", () => {
     await drain(stream);
     const spans = otel.getFinishedSpans();
     const compaction = requireSpan(spans, "apex.context.compact");
-    const summary = spans.find(
-      (span) => span.name === "ai.generateText.doGenerate",
+    const summary = summarySpan(spans);
+    const resumed = spans.find(
+      (span) =>
+        span.name === "ai.streamText" &&
+        span.attributes["ai.telemetry.functionId"] !== SUMMARIZE_FN,
     );
-    const resumed = requireSpan(spans, "ai.streamText");
+    if (!resumed) throw new Error("no resumed agent-turn span");
     expect(compaction?.attributes).toMatchObject({
       "apex.compaction.outcome": "completed",
       "pensar.session.id": "ses_child",
@@ -387,7 +410,7 @@ describe("compaction observability", () => {
     ).toBe(true);
   });
 
-  it("preserves tool evidence omitted by the summarizer and links its actual request and response", async () => {
+  it("passes tool evidence to the summarizer and links its actual request and response", async () => {
     const summaryModel = model();
     state.model = summaryModel;
     const messages: ModelMessage[] = [
@@ -418,11 +441,12 @@ describe("compaction observability", () => {
       prompt: "original task",
       system: "research system",
     });
-    const summary = requireSpan(spans, "ai.generateText.doGenerate");
+    const summary = summarySpan(spans);
+    if (!summary) throw new Error("no summarizer model span");
     expect(summary.attributes["ai.prompt.messages"]).toContain(
       "continue researching",
     );
-    expect(summary.attributes["ai.prompt.messages"]).not.toContain(
+    expect(summary.attributes["ai.prompt.messages"]).toContain(
       "evidence evidence",
     );
     expect(summary.attributes["ai.response.text"]).toBe("A compact summary");
@@ -433,7 +457,8 @@ describe("compaction observability", () => {
       String(compact.attributes["apex.compaction.evidence.after.json"]),
     );
     expect(after.messages[0].content).toContain("A compact summary");
-    expect(summaryModel.doStreamCalls[0]?.prompt).toMatchObject([
+    // [0] is the summarization call; the resumed agent turn follows it.
+    expect(summaryModel.doStreamCalls[1]?.prompt).toMatchObject([
       { role: "system", content: after.system },
       {
         role: "user",
@@ -538,11 +563,30 @@ describe("compaction observability", () => {
     }
   });
 
+  it("summarizes over a stream, so a slow summary is never a silent connection", async () => {
+    // A one-shot summarization returns no response headers until the whole summary
+    // exists, which anything watching time-to-headers reads as a dead connection.
+    // Compaction became unreachable that way, so the call must stream.
+    const summaryModel = model();
+    state.model = summaryModel;
+    await drain(
+      createSummarizationStream(
+        [{ role: "user", content: "history" }],
+        { prompt: "continue", model: MODEL, silent: true },
+        summaryModel,
+      ),
+    );
+    expect(summaryModel.doGenerateCalls).toHaveLength(0);
+    expect(JSON.stringify(summaryModel.doStreamCalls[0]?.prompt)).toContain(
+      "Compact this conversation",
+    );
+  });
+
   it("records summary cancellation without hiding the original error", async () => {
     const cancelled = new Error("cancelled");
     cancelled.name = "AbortError";
     state.model = new MockLanguageModelV3({
-      doGenerate: async () => {
+      doStream: async () => {
         throw cancelled;
       },
     });
@@ -569,10 +613,10 @@ describe("compaction observability", () => {
     const normal = model();
     state.model = new MockLanguageModelV3({
       modelId: MODEL,
-      doGenerate: async () => {
-        throw new Error("context length exceeded");
-      },
-      doStream: (options) => normal.doStream(options),
+      doStream: (options) =>
+        isSummaryCall(options)
+          ? Promise.reject(new Error("context length exceeded"))
+          : normal.doStream(options),
     });
     await runWithAiPayloadCapture(true, () =>
       drain(
@@ -612,10 +656,18 @@ describe("compaction observability", () => {
         String(reset?.attributes["apex.compaction.evidence.after.json"]),
       )?.messages,
     ).toEqual([{ role: "user", content: "original task" }]);
+    // A resumed agent turn must carry the reset's span id. Summarizer attempts are
+    // `ai.streamText` spans too now, so they are excluded rather than assumed last.
     expect(
-      spans.find((span) => span.name === "ai.streamText")?.attributes[
-        "ai.telemetry.metadata.compactionSpanId"
-      ],
-    ).toBe(reset?.spanContext().spanId);
+      spans
+        .filter(
+          (span) =>
+            span.name === "ai.streamText" &&
+            span.attributes["ai.telemetry.functionId"] !== SUMMARIZE_FN,
+        )
+        .map(
+          (span) => span.attributes["ai.telemetry.metadata.compactionSpanId"],
+        ),
+    ).toContain(reset?.spanContext().spanId);
   });
 });

package/src/core/ai/index.ts

@@ -69,5 +69,16 @@ export {
   stringifyCanonicalJson,
   toJsonValue,
 } from "./native-rollout-evidence";
+export {
+  getModelSettings,
+  getReasoningCapabilities,
+  getReasoningChoices,
+  REASONING_SETTINGS,
+  type ReasoningCapabilities,
+  type ReasoningEffort,
+  type ReasoningSetting,
+  resolveReasoningSetting,
+  validateReasoningSetting,
+} from "./reasoning";
 export type { AIAuthConfig } from "./utils";
 export { buildAuthConfig } from "./utils";

package/src/core/ai/models/anthropic.ts

@@ -112,6 +112,12 @@ export const ANTHROPIC_MODELS: ModelInfo[] = [
     provider: "anthropic",
     contextLength: 1000000,
   },
+  {
+    id: "claude-haiku-5-5",
+    name: "Claude Haiku 5.5",
+    provider: "anthropic",
+    contextLength: 1000000,
+  },
   {
     id: "claude-sonnet-5-5",
     name: "Claude Sonnet 5.5",

package/src/core/ai/models/bedrock.ts

@@ -465,6 +465,18 @@ export const BEDROCK_MODELS: ModelInfo[] = [
     provider: "bedrock",
     contextLength: 262000,
   },
+  {
+    id: "anthropic.claude-haiku-5-5",
+    name: "Claude Haiku 5.5 (Bedrock)",
+    provider: "bedrock",
+    contextLength: 1000000,
+  },
+  {
+    id: "us.anthropic.claude-haiku-5-5",
+    name: "Claude Haiku 5.5 (US)",
+    provider: "bedrock",
+    contextLength: 1000000,
+  },
   {
     id: "anthropic.claude-opus-4-8",
     name: "Claude Opus 4.8 (Bedrock)",
@@ -597,4 +609,10 @@ export const BEDROCK_MODELS: ModelInfo[] = [
     provider: "bedrock",
     contextLength: 1000000,
   },
+  {
+    id: "global.anthropic.claude-haiku-5-5",
+    name: "Claude Haiku 5.5 (Global)",
+    provider: "bedrock",
+    contextLength: 1000000,
+  },
 ];

package/src/core/ai/models/claude-capabilities.ts

@@ -1,22 +1,43 @@
-type ClaudeCapabilities = {
+export type ClaudeThinkingEffort = "low" | "medium" | "high" | "xhigh" | "max";
+
+export type ClaudeCapabilities = {
   alwaysOnThinking: boolean;
   bindsThinking?: boolean;
+  defaultEffort: ClaudeThinkingEffort;
 };
 
 const CLAUDE_CAPABILITIES: Record<string, ClaudeCapabilities> = {
-  "claude-sonnet-5-5": { alwaysOnThinking: true, bindsThinking: true },
-  "claude-opus-5-5": { alwaysOnThinking: true, bindsThinking: true },
-  "claude-fable-5-1": { alwaysOnThinking: true, bindsThinking: true },
-  "claude-opus-5": { alwaysOnThinking: false },
-  "claude-sonnet-5": { alwaysOnThinking: false },
-  "claude-fable-5": { alwaysOnThinking: true },
+  "claude-haiku-5-5": {
+    alwaysOnThinking: false,
+    bindsThinking: true,
+    defaultEffort: "medium",
+  },
+  "claude-sonnet-5-5": {
+    alwaysOnThinking: true,
+    bindsThinking: true,
+    defaultEffort: "high",
+  },
+  "claude-opus-5-5": {
+    alwaysOnThinking: true,
+    bindsThinking: true,
+    defaultEffort: "medium",
+  },
+  "claude-fable-5-1": {
+    alwaysOnThinking: true,
+    bindsThinking: true,
+    defaultEffort: "high",
+  },
+  "claude-opus-5": { alwaysOnThinking: false, defaultEffort: "high" },
+  "claude-sonnet-5": { alwaysOnThinking: false, defaultEffort: "high" },
+  "claude-fable-5": { alwaysOnThinking: true, defaultEffort: "high" },
 };
 
 export function getClaudeCapabilities(
   modelId: string,
 ): ClaudeCapabilities | undefined {
   const nativeId = modelId
-    .replace(/^(concentrate:|anthropic\/)/, "")
+    .replace(/^(concentrate:|pensar:)/, "")
+    .replace(/^(?:(?:us|global|eu|au|jp)\.)?anthropic[./]/, "")
     .replace(/(\d)\.(\d)/g, "$1-$2");
   return CLAUDE_CAPABILITIES[nativeId];
 }

package/src/core/ai/models/concentrate.ts

@@ -1,5 +1,15 @@
 import type { ModelInfo } from "../ai";
 
+// Concentrate does not route on max_output_tokens, so these use the lowest
+// ceiling among each slug's upstream providers.
+export const CONCENTRATE_MAX_OUTPUT_TOKENS: Readonly<Record<string, number>> = {
+  "concentrate:kimi-k3": 32_768,
+  "concentrate:qwen3.8-max": 32_768,
+  "concentrate:qwen3.8-27b": 32_768,
+  "concentrate:deepseek-v4-1-flash": 32_768,
+  "concentrate:deepseek-v4-1-flash-1017": 131_072,
+};
+
 // Curated from Concentrate's public model catalog. The local prefix keeps
 // gateway selections distinct from direct-provider models and is not sent.
 export const CONCENTRATE_MODELS: ModelInfo[] = [
@@ -153,6 +163,36 @@ export const CONCENTRATE_MODELS: ModelInfo[] = [
     provider: "concentrate",
     contextLength: 1000000,
   },
+  {
+    id: "concentrate:qwen3.8-max",
+    name: "Qwen3.8 Max",
+    provider: "concentrate",
+    contextLength: 256000,
+  },
+  {
+    id: "concentrate:qwen3.8-27b",
+    name: "Qwen3.8 27B",
+    provider: "concentrate",
+    contextLength: 262144,
+  },
+  {
+    id: "concentrate:deepseek-v4-1-flash-1017",
+    name: "DeepSeek V4.1 Flash 1017",
+    provider: "concentrate",
+    contextLength: 1048576,
+  },
+  {
+    id: "concentrate:deepseek-v4-1-flash",
+    name: "DeepSeek V4.1 Flash",
+    provider: "concentrate",
+    contextLength: 1000000,
+  },
+  {
+    id: "concentrate:deepseek-v4-pro-0813",
+    name: "DeepSeek V4 Pro 0813",
+    provider: "concentrate",
+    contextLength: 1048576,
+  },
   {
     id: "concentrate:deepseek-v3-2",
     name: "DeepSeek V3.2",
@@ -165,6 +205,12 @@ export const CONCENTRATE_MODELS: ModelInfo[] = [
     provider: "concentrate",
     contextLength: 256000,
   },
+  {
+    id: "concentrate:kimi-k3",
+    name: "Kimi K3",
+    provider: "concentrate",
+    contextLength: 1048576,
+  },
   {
     id: "concentrate:kimi-k2-thinking",
     name: "Kimi K2 Thinking",
@@ -177,6 +223,12 @@ export const CONCENTRATE_MODELS: ModelInfo[] = [
     provider: "concentrate",
     contextLength: 204800,
   },
+  {
+    id: "concentrate:mimo-v2.6-pro",
+    name: "MiMo V2.6 Pro",
+    provider: "concentrate",
+    contextLength: 1048576,
+  },
   {
     id: "concentrate:glm-5.3",
     name: "GLM-5.3",

package/src/core/ai/models/index.ts

@@ -12,9 +12,15 @@ import { ANTHROPIC_MODELS } from "./anthropic";
 import { BEDROCK_MODELS } from "./bedrock";
 import { getClaudeCapabilities } from "./claude-capabilities";
 
-export { getClaudeCapabilities } from "./claude-capabilities";
+export {
+  type ClaudeThinkingEffort,
+  getClaudeCapabilities,
+} from "./claude-capabilities";
 
-import { CONCENTRATE_MODELS } from "./concentrate";
+import {
+  CONCENTRATE_MAX_OUTPUT_TOKENS,
+  CONCENTRATE_MODELS,
+} from "./concentrate";
 import { GOOGLE_MODELS } from "./google";
 import { INCEPTION_MODELS } from "./inception";
 // Bedrock Mantle (OpenAI Responses API) — curated manually.
@@ -100,6 +106,8 @@ export function getMaxOutputTokens(
 function lookupOutputBudgetByPattern(modelId: string): number {
   if (modelId === "chat-latest") return 128_000;
   if (getClaudeCapabilities(modelId)) return 128_000;
+  const concentrateBudget = CONCENTRATE_MAX_OUTPUT_TOKENS[modelId];
+  if (concentrateBudget) return concentrateBudget;
   // OpenRouter uses dots in Claude version numbers (anthropic/claude-opus-4.6)
   // while native Anthropic uses dashes (claude-opus-4-6-20250929). Normalize
   // digit.digit sequences to dashes so all Claude patterns match both forms.
@@ -219,14 +227,22 @@ function lookupOutputBudgetByPattern(modelId: string): number {
     return 131_072;
   }
 
+  if (modelId.includes("mimo-v2.6-pro")) {
+    return 131_072;
+  }
+
+  if (modelId.includes("kimi-k3")) {
+    return 131_072;
+  }
+
   // GLM 5 models ship a ~131K (128Ki) max-output window. Matching the family
   // keeps Bedrock and OpenRouter model IDs on the same budget.
   if (modelId.includes("glm-5")) {
     return 131_072;
   }
 
-  // V4.1 Flash's larger output window must precede the legacy DeepSeek fallback.
-  if (modelId.includes("deepseek-v4.1-flash")) {
+  // DeepSeek V4's larger output window must precede the legacy DeepSeek fallback.
+  if (/deepseek-v4[.-]/.test(modelId)) {
     return 384_000;
   }
 
@@ -235,6 +251,11 @@ function lookupOutputBudgetByPattern(modelId: string): number {
     return 8_192;
   }
 
+  // Qwen3.8's first-party 131K output window must precede the Qwen fallback.
+  if (modelId.includes("qwen3.8")) {
+    return 131_072;
+  }
+
   // Qwen3 Coder ships a 16K max-output window over its 128K+ context. Match the
   // whole `qwen` family so future Bedrock/OpenRouter qwen ids don't silently
   // inherit the 4,096 default.

package/src/core/ai/models/models.test.ts

@@ -103,6 +103,11 @@ describe("getMaxOutputTokens", () => {
   });
 
   it("recognizes Claude tier-specific budgets", () => {
+    expect(getMaxOutputTokens("claude-haiku-5-5")).toBe(128_000);
+    expect(getMaxOutputTokens("global.anthropic.claude-haiku-5-5")).toBe(
+      128_000,
+    );
+    expect(getMaxOutputTokens("anthropic/claude-haiku-5.5")).toBe(128_000);
     expect(getMaxOutputTokens("claude-sonnet-4-5-20250929")).toBe(64_000);
     expect(getMaxOutputTokens("claude-opus-4-5-20250101")).toBe(64_000);
     expect(getMaxOutputTokens("claude-haiku-4-5-20251001")).toBe(64_000);
@@ -111,6 +116,24 @@ describe("getMaxOutputTokens", () => {
     expect(getMaxOutputTokens("claude-3-5-haiku-20241022")).toBe(8_192);
   });
 
+  it("registers Claude Haiku 5.5 across Anthropic, Bedrock, and OpenRouter", () => {
+    expect(getModelInfo("claude-haiku-5-5")).toMatchObject({
+      name: "Claude Haiku 5.5",
+      provider: "anthropic",
+      contextLength: 1_000_000,
+    });
+    expect(getModelInfo("global.anthropic.claude-haiku-5-5")).toMatchObject({
+      name: "Claude Haiku 5.5 (Global)",
+      provider: "bedrock",
+      contextLength: 1_000_000,
+    });
+    expect(getModelInfo("anthropic/claude-haiku-5.5")).toMatchObject({
+      name: "Claude Haiku 5.5",
+      provider: "openrouter",
+      contextLength: 1_000_000,
+    });
+  });
+
   it("pins latest-tier Claude (4.6 + 4.7 + 4.8) to 128K output", () => {
     // These ship the 128K window; they MUST not fall through to the
     // generic `claude-opus-4-` / `claude-sonnet-4-` catch-alls (32K/64K)
@@ -261,7 +284,6 @@ describe("OpenAI reasoning effort support", () => {
       "high",
       "xhigh",
       "max",
-      "ultra",
     ]);
     expect(modelSupportsOpenAIReasoning("gpt-5.6-terra")).toBe(true);
     expect(modelSupportsOpenAIReasoning("gpt-5.6-luna")).toBe(true);
@@ -274,12 +296,31 @@ describe("OpenAI reasoning effort support", () => {
     expect(modelSupportsOpenAIReasoning("o3-mini")).toBe(true);
   });
 
-  it("does not apply reasoning effort to unsupported GPT-5 variants", () => {
-    expect(modelSupportsOpenAIReasoning("gpt-5-nano")).toBe(false);
-    expect(modelSupportsOpenAIReasoning("gpt-5.1-codex-mini")).toBe(false);
-    expect(modelSupportsOpenAIReasoning("gpt-5.1-codex-max")).toBe(false);
-    expect(modelSupportsOpenAIReasoning("gpt-5.3-codex")).toBe(false);
-    expect(normalizeOpenAIReasoningEffort("gpt-5-nano")).toBeUndefined();
+  it("uses verified effort support for smaller and Codex models", () => {
+    expect(getOpenAIReasoningEfforts("gpt-5-nano")).toEqual([
+      "minimal",
+      "low",
+      "medium",
+      "high",
+    ]);
+    expect(getOpenAIReasoningEfforts("gpt-5.1-codex-mini")).toEqual([
+      "low",
+      "medium",
+      "high",
+    ]);
+    expect(getOpenAIReasoningEfforts("gpt-5.1-codex-max")).toEqual([
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+    ]);
+    expect(getOpenAIReasoningEfforts("gpt-5.3-codex")).toEqual([
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+    ]);
+    expect(modelSupportsOpenAIReasoning("gpt-future-99")).toBe(false);
   });
 
   it("defaults only supported reasoning models to medium", () => {
@@ -288,3 +329,134 @@ describe("OpenAI reasoning effort support", () => {
     );
   });
 });
+
+describe("open-weight gateway models", () => {
+  it.each([
+    {
+      id: "xiaomi/mimo-v2.6-pro",
+      name: "MiMo-V2.6-Pro",
+      provider: "openrouter",
+      contextLength: 1_050_000,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "concentrate:mimo-v2.6-pro",
+      name: "MiMo V2.6 Pro",
+      provider: "concentrate",
+      contextLength: 1_048_576,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "moonshotai/kimi-k3",
+      name: "Kimi K3",
+      provider: "openrouter",
+      contextLength: 1_048_576,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "concentrate:kimi-k3",
+      name: "Kimi K3",
+      provider: "concentrate",
+      contextLength: 1_048_576,
+      maxOutputTokens: 32_768,
+    },
+    {
+      id: "qwen/qwen3.8-max-0902",
+      name: "Qwen3.8 Max (0902)",
+      provider: "openrouter",
+      contextLength: 1_000_000,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "qwen/qwen3.8-27b",
+      name: "Qwen3.8 27B",
+      provider: "openrouter",
+      contextLength: 1_000_000,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "qwen/qwen3.8-flash",
+      name: "Qwen3.8 Flash",
+      provider: "openrouter",
+      contextLength: 1_000_000,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "concentrate:qwen3.8-max",
+      name: "Qwen3.8 Max",
+      provider: "concentrate",
+      contextLength: 256_000,
+      maxOutputTokens: 32_768,
+    },
+    {
+      id: "concentrate:qwen3.8-27b",
+      name: "Qwen3.8 27B",
+      provider: "concentrate",
+      contextLength: 262_144,
+      maxOutputTokens: 32_768,
+    },
+    {
+      id: "deepseek/deepseek-v4.1-flash-1017",
+      name: "DeepSeek V4.1 Flash 1017",
+      provider: "openrouter",
+      contextLength: 1_048_576,
+      maxOutputTokens: 384_000,
+    },
+    {
+      id: "deepseek/deepseek-v4-pro-0813",
+      name: "DeepSeek V4 Pro 0813",
+      provider: "openrouter",
+      contextLength: 1_048_576,
+      maxOutputTokens: 384_000,
+    },
+    {
+      id: "concentrate:deepseek-v4-1-flash",
+      name: "DeepSeek V4.1 Flash",
+      provider: "concentrate",
+      contextLength: 1_000_000,
+      maxOutputTokens: 32_768,
+    },
+    {
+      id: "concentrate:deepseek-v4-1-flash-1017",
+      name: "DeepSeek V4.1 Flash 1017",
+      provider: "concentrate",
+      contextLength: 1_048_576,
+      maxOutputTokens: 131_072,
+    },
+    {
+      id: "concentrate:deepseek-v4-pro-0813",
+      name: "DeepSeek V4 Pro 0813",
+      provider: "concentrate",
+      contextLength: 1_048_576,
+      maxOutputTokens: 384_000,
+    },
+  ])("registers $id with its gateway limits", ({
+    id,
+    maxOutputTokens,
+    ...info
+  }) => {
+    expect(getModelInfo(id)).toMatchObject(info);
+    expect(getMaxOutputTokens(id)).toBe(maxOutputTokens);
+  });
+
+  it.each([
+    "concentrate:mimo-v2.6-pro",
+    "concentrate:kimi-k3",
+    "concentrate:qwen3.8-max",
+    "concentrate:qwen3.8-27b",
+    "concentrate:deepseek-v4-1-flash",
+    "concentrate:deepseek-v4-1-flash-1017",
+    "concentrate:deepseek-v4-pro-0813",
+  ])("offers every effort Concentrate can route for %s", (id) => {
+    expect(modelSupportsOpenAIReasoning(id)).toBe(true);
+    expect(getOpenAIReasoningEfforts(id)).toEqual([
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+      "max",
+    ]);
+    expect(normalizeOpenAIReasoningEffort(id)).toBe("medium");
+    expect(normalizeOpenAIReasoningEffort(id, "max")).toBe("max");
+  });
+});