@vellumai/assistant 0.10.12-staging.2 → 0.10.12

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vellumai/assistant",
-  "version": "0.10.12-staging.2",
+  "version": "0.10.12",
   "license": "MIT",
   "type": "module",
   "exports": {

package/src/__tests__/config-loader-backfill.test.ts

@@ -855,10 +855,9 @@ describe("loadConfig startup behavior", () => {
     expect(effective.balanced?.model).toBe("accounts/fireworks/models/glm-5p2");
     expect(effective.balanced?.effort).toBe("high");
     expect(effective.balanced?.source).toBe("managed");
-    // Quality pins Anthropic Fable, the most capable managed model.
+    // Quality pins OpenAI Sol, the most capable managed model.
     expect(effective["quality-optimized"]?.provider).toBe("vellum");
-    expect(effective["quality-optimized"]?.model).toBe("claude-fable-5");
-    expect(effective["quality-optimized"]?.model).toBe("claude-fable-5");
+    expect(effective["quality-optimized"]?.model).toBe("gpt-5.6-sol");
     // Speed is served by DeepSeek V4 Flash on Fireworks.
     expect(effective["cost-optimized"]?.provider).toBe("vellum");
     expect(effective["cost-optimized"]?.model).toBe(
@@ -916,7 +915,7 @@ describe("loadConfig startup behavior", () => {
     const raw = JSON.parse(readFileSync(CONFIG_PATH, "utf-8"));
 
     expect(raw.llm.profiles["quality-optimized"]).toEqual(legacy);
-    // The shadow wins over the catalog body (which serves claude-fable-5).
+    // The shadow wins over the catalog body (which serves gpt-5.6-sol).
     const effectiveQuality = getEffectiveProfile(
       raw.llm.profiles,
       "quality-optimized",
@@ -941,7 +940,7 @@ describe("loadConfig startup behavior", () => {
       "quality-optimized",
     );
     expect(effectiveQuality?.source).toBe("managed");
-    expect(effectiveQuality?.model).toBe("claude-fable-5");
+    expect(effectiveQuality?.model).toBe("gpt-5.6-sol");
   });
 
   test("platform boot leaves a drifted managed entry byte-identical; resolution serves catalog content", () => {

package/src/config/__tests__/default-profile-catalog.test.ts

@@ -53,6 +53,15 @@ describe("getEffectiveProfiles", () => {
     }
   });
 
+  test("the managed Quality profile routes GPT-5.6 Sol through OpenAI", () => {
+    const quality = CODE_DEFAULT_PROFILE_ENTRIES["quality-optimized"];
+    expect(quality.model).toBe("gpt-5.6-sol");
+    expect(resolveRoutingIdentity(quality.provider, quality.model)).toEqual({
+      connectionName: "vellum",
+      expectedProvider: "openai",
+    });
+  });
+
   test("defaults absent from the workspace resolve from the catalog; os-beta stays flag-gated", () => {
     const effective = getEffectiveProfiles(undefined);
     expect(Object.keys(effective).sort()).toEqual(

package/src/config/default-profile-catalog.ts

@@ -79,7 +79,7 @@ const VELLUM_PROFILE_IMPLS: Record<ProfileMatrixKey, DefaultProfileTemplate> = {
     },
   },
   "quality-optimized": {
-    model: "claude-fable-5",
+    model: "gpt-5.6-sol",
     provider: "vellum",
     source: "managed",
     label: "Quality",

package/src/providers/anthropic/client.ts

@@ -973,17 +973,19 @@ export class AnthropicProvider implements Provider {
         (restConfig as Record<string, unknown>).model?.toString() ?? this.model;
       const isHaiku = effectiveModel.includes("haiku");
       const supportsEffort = !isHaiku;
-      // opus-4-7 / opus-4-8 and sonnet-5 reject `temperature`, `top_p`, and
-      // `top_k` with a 400 "`temperature`/`top_p` is deprecated for this model"
-      // — model-wide, not effort-conditional (verified 2026-06-23). opus-4-6 /
-      // sonnet-4-6 / haiku-4-5 still accept them. fable-5 is included
-      // conservatively (a frontier model that could not be verified directly
-      // but follows the same deprecation direction). Stripping the params here
-      // keeps callers that set them (e.g. the memory-v3 L2 selector's
-      // `temperature: 0`) from 400ing. OpenRouter `anthropic/...` models
-      // delegate to this provider, so the bare-id suffix is what matches.
+      // opus-4-7 / opus-4-8 / opus-5 and sonnet-5 reject `temperature`,
+      // `top_p`, and `top_k` with a 400 "`temperature`/`top_p` is deprecated
+      // for this model" — model-wide, not effort-conditional (verified
+      // 2026-06-23). opus-4-6 / sonnet-4-6 / haiku-4-5 still accept them.
+      // fable-5 is included conservatively (a frontier model that could not be
+      // verified directly but follows the same deprecation direction).
+      // Stripping the params here keeps callers that set them (e.g. the
+      // memory-v3 L2 selector's `temperature: 0`) from 400ing. OpenRouter
+      // `anthropic/...` models delegate to this provider, so the bare-id
+      // suffix is what matches.
       const deprecatesSamplingParams =
         /claude-opus-4-[78]\b/.test(effectiveModel) ||
+        /claude-opus-5\b/.test(effectiveModel) ||
         /claude-sonnet-5\b/.test(effectiveModel) ||
         effectiveModel.startsWith("claude-fable-");
       const mergedOutputConfig = {

package/src/providers/model-catalog.ts

@@ -179,6 +179,23 @@ const RAW_PROVIDER_CATALOG: ProviderCatalogEntry[] = [
           cacheReadPer1mTokens: 1,
         },
       },
+      {
+        id: "claude-opus-5",
+        displayName: "Claude Opus 5",
+        contextWindowTokens: 1000000,
+        maxOutputTokens: 128000,
+        longContextPricingThresholdTokens: 200000,
+        supportsThinking: true,
+        supportsCaching: true,
+        supportsVision: true,
+        supportsToolUse: true,
+        pricing: {
+          inputPer1mTokens: 5,
+          outputPer1mTokens: 25,
+          cacheWritePer1mTokens: 6.25,
+          cacheReadPer1mTokens: 0.5,
+        },
+      },
       {
         id: "claude-opus-4-8",
         displayName: "Claude Opus 4.8",
@@ -1000,6 +1017,23 @@ const RAW_PROVIDER_CATALOG: ProviderCatalogEntry[] = [
           cacheReadPer1mTokens: 1,
         },
       },
+      {
+        id: "anthropic/claude-opus-5",
+        displayName: "Claude Opus 5",
+        contextWindowTokens: 1000000,
+        maxOutputTokens: 128000,
+        longContextPricingThresholdTokens: 200000,
+        supportsThinking: true,
+        supportsCaching: true,
+        supportsVision: true,
+        supportsToolUse: true,
+        pricing: {
+          inputPer1mTokens: 5,
+          outputPer1mTokens: 25,
+          cacheWritePer1mTokens: 6.25,
+          cacheReadPer1mTokens: 0.5,
+        },
+      },
       {
         id: "anthropic/claude-opus-4.8",
         displayName: "Claude Opus 4.8",
@@ -1733,6 +1767,23 @@ const RAW_PROVIDER_CATALOG: ProviderCatalogEntry[] = [
           cacheReadPer1mTokens: 1,
         },
       },
+      {
+        id: "anthropic/claude-opus-5",
+        displayName: "Claude Opus 5",
+        contextWindowTokens: 1000000,
+        maxOutputTokens: 128000,
+        longContextPricingThresholdTokens: 200000,
+        supportsThinking: true,
+        supportsCaching: true,
+        supportsVision: true,
+        supportsToolUse: true,
+        pricing: {
+          inputPer1mTokens: 5,
+          outputPer1mTokens: 25,
+          cacheWritePer1mTokens: 6.25,
+          cacheReadPer1mTokens: 0.5,
+        },
+      },
       {
         id: "anthropic/claude-opus-4.8",
         displayName: "Claude Opus 4.8",