feat(providers): refresh Fable, Cursor, and Devin catalogs (#12367)

Validado em lote numa worktree combinada com #12524, #12538, #12277 e #12367 sobre o tip de release/v3.8.51: os quatro boardaram sem conflito (áreas disjuntas — zai-web, nvidia, clova, cursor/devin/fable). typecheck:core limpo, check:provider-consistency OK (272 entradas REGISTRY, 355 providers canônicos), check:known-symbols OK, e 305/305 nos testes tocados pelos quatro PRs. Os IDs de modelo adicionados foram conferidos individualmente. Obrigado, @backryun.
2026-09-18 12:52:25 +03:00 · 2026-09-03 19:50:41 +09:00
parent 82c64d76d3
commit 6795783228
42 changed files with 1835 additions and 921 deletions
--- a/src/shared/constants/cliTools.ts
+++ b/src/shared/constants/cliTools.ts
@@ -45,7 +45,7 @@ export const CLI_TOOLS: Record<string, CliCatalogEntry> = {
        name: "Claude Fable",
        alias: "fable",
        envKey: "ANTHROPIC_DEFAULT_FABLE_MODEL",
-        defaultValue: _cc.fable ? `cc/${_cc.fable}` : "cc/claude-fable-5",
+        defaultValue: _cc.fable ? `cc/${_cc.fable}` : "cc/claude-fable-5-1",
        isTopLevel: true,
      },
      {
--- a/src/shared/constants/modelSpecs.ts
+++ b/src/shared/constants/modelSpecs.ts
@@ -24,9 +24,13 @@ export interface ModelSpec {
  // Model ONLY supports adaptive thinking: manual extended thinking was removed. Sending
  // `thinking.type:"enabled"` or any `thinking.budget_tokens` returns HTTP 400; reasoning
  // is steered exclusively by `output_config.effort` (low/medium/high/xhigh/max). True for
-  // Claude Opus 4.7 and later (Opus 4.7/4.8/5, Fable 5). Per Anthropic's migration guide,
+  // Claude Opus 4.7 and later (Opus 4.7/4.8/5, Fable 5/5.1). Per Anthropic's migration guide,
  // any request that tries to set a fixed thinking budget gets a 400 error.
  adaptiveThinkingOnly?: boolean;
+  // The model rejects tool_choice values that require a tool call. Keep tools available,
+  // but normalize a forced choice to the default auto behavior before dispatch. Fable 5.1 always runs
+  // adaptive thinking, so forced tool use cannot be combined with any valid request.
+  rejectsForcedToolChoice?: boolean;
  // Highest effort accepted while `thinking.type:"disabled"` is present. Claude Opus 5
  // rejects disabled thinking with xhigh/max, while accepting it through high.
  maxEffortWhenThinkingDisabled?: "high";
@@ -371,6 +375,21 @@ export const MODEL_SPECS: Record<string, ModelSpec> = {
    aliases: BEDROCK_CLAUDE_ALIASES("claude-opus-4-7", "claude-opus-4.7"),
  },

+  // ── Claude Fable 5.1 ────────────────────────────────────────────
+  "claude-fable-5-1": {
+    maxOutputTokens: 128000,
+    contextWindow: 1000000,
+    defaultThinkingBudget: 32000,
+    thinkingBudgetCap: 120000,
+    supportsThinking: true,
+    supportsTools: true,
+    supportsVision: true,
+    rejectsThinkingDisabled: true,
+    adaptiveThinkingOnly: true,
+    rejectsForcedToolChoice: true,
+    aliases: BEDROCK_CLAUDE_ALIASES("claude-fable-5-1"),
+  },
+
  // ── Claude Fable 5 ──────────────────────────────────────────────
  "claude-fable-5": {
    maxOutputTokens: 128000,
@@ -849,9 +868,39 @@ export function normalizeThinkingForModel<T extends Record<string, unknown>>(
    getModelSpec(modelId)?.rejectsThinkingDisabled
  ) {
    const { thinking: _omitted, ...rest } = body as Record<string, unknown>;
-    return rest as T;
+    return normalizeForcedToolChoiceForModel(rest as T, modelId);
  }
-  return body;
+  return normalizeForcedToolChoiceForModel(body, modelId);
+}
+
+/**
+ * Normalize tool-choice constraints that a resolved model cannot accept.
+ *
+ * Claude Fable 5.1 always uses adaptive thinking and rejects tool choices that force
+ * either any tool or one named tool. Preserve the declared tools and every unrelated
+ * request field, but drop the choice to select the default `auto` behavior so routing a
+ * request to Fable 5.1 does not turn a recoverable preference into an upstream 400.
+ */
+export function normalizeForcedToolChoiceForModel<T extends Record<string, unknown>>(
+  body: T,
+  modelId: string
+): T {
+  if (!getModelSpec(modelId)?.rejectsForcedToolChoice) return body;
+
+  const toolChoice = body.tool_choice;
+  const forced =
+    toolChoice === "required" ||
+    toolChoice === "any" ||
+    (toolChoice !== null &&
+      typeof toolChoice === "object" &&
+      !Array.isArray(toolChoice) &&
+      ["any", "tool", "function"].includes(
+        String((toolChoice as Record<string, unknown>).type || "").toLowerCase()
+      ));
+  if (!forced) return body;
+
+  const { tool_choice: _omitted, ...rest } = body;
+  return rest as T;
 }

 export function capMaxOutputTokens(modelId: string, requested?: number): number | undefined {
--- a/src/shared/constants/pricing/default-pricing.ts
+++ b/src/shared/constants/pricing/default-pricing.ts
@@ -7,10 +7,12 @@ import { DEFAULT_PRICING_OAUTH } from "./oauth-subscriptions";
 import { DEFAULT_PRICING_FRONTIER } from "./frontier-labs";
 import { DEFAULT_PRICING_INFERENCE } from "./inference-hosts";
 import { DEFAULT_PRICING_REGIONAL } from "./regional";
+import { DEFAULT_PRICING_DEVIN } from "./devin";

 export const DEFAULT_PRICING = {
  ...DEFAULT_PRICING_OAUTH,
  ...DEFAULT_PRICING_FRONTIER,
  ...DEFAULT_PRICING_INFERENCE,
  ...DEFAULT_PRICING_REGIONAL,
+  ...DEFAULT_PRICING_DEVIN,
 };
--- a/src/shared/constants/pricing/devin.ts
+++ b/src/shared/constants/pricing/devin.ts
@@ -0,0 +1,142 @@
+type DevinTokenPricing = {
+  input: number;
+  cached: number;
+  output: number;
+};
+
+const QUALITY_EFFORTS = ["max", "xhigh", "high", "medium", "low"] as const;
+const GPT_EFFORTS = ["max", "xhigh", "high", "medium", "low", "none"] as const;
+
+function variantIds(base: string, efforts: readonly string[]): string[] {
+  return efforts.map((effort) => `${base}-${effort}`);
+}
+
+function fastVariantIds(base: string): string[] {
+  return QUALITY_EFFORTS.map((effort) => `${base}-${effort}-fast`);
+}
+
+function priorityVariantIds(base: string): string[] {
+  return GPT_EFFORTS.map((effort) => `${base}-${effort}-priority`);
+}
+
+function priced(ids: readonly string[], pricing: DevinTokenPricing) {
+  return Object.fromEntries(ids.map((id) => [id, pricing]));
+}
+
+const CLAUDE_FABLE_5_1 = { input: 10, cached: 0.25, output: 50 };
+const CLAUDE_OPUS = { input: 5, cached: 0.5, output: 25 };
+const CLAUDE_OPUS_FAST = { input: 10, cached: 1, output: 50 };
+const CLAUDE_SONNET_5 = { input: 2, cached: 0.2, output: 10 };
+const CLAUDE_SONNET_4_6 = { input: 3, cached: 0.3, output: 15 };
+const CLAUDE_HAIKU_4_5 = { input: 1, cached: 0.1, output: 5 };
+
+const GPT_5_6_SOL = { input: 4, cached: 0.4, output: 20 };
+const GPT_5_6_SOL_FAST = { input: 8, cached: 0.8, output: 40 };
+const GPT_5_6_TERRA = { input: 2, cached: 0.2, output: 12 };
+const GPT_5_6_TERRA_FAST = { input: 4, cached: 0.4, output: 24 };
+const GPT_5_6_LUNA = { input: 0.2, cached: 0.02, output: 1.2 };
+const GPT_5_6_LUNA_FAST = { input: 0.4, cached: 0.04, output: 2.4 };
+
+/**
+ * Exact per-UID rates returned by Devin's authenticated live catalog on
+ * 2026-09-02. Rates are USD per one million tokens.
+ */
+export const DEVIN_MODEL_PRICING: Record<string, DevinTokenPricing> = {
+  ...priced(variantIds("claude-fable-5-1", QUALITY_EFFORTS), CLAUDE_FABLE_5_1),
+  ...priced(variantIds("claude-opus-5", QUALITY_EFFORTS), CLAUDE_OPUS),
+  ...priced(fastVariantIds("claude-opus-5"), CLAUDE_OPUS_FAST),
+  ...priced(variantIds("claude-opus-4-8", QUALITY_EFFORTS), CLAUDE_OPUS),
+  ...priced(fastVariantIds("claude-opus-4-8"), CLAUDE_OPUS_FAST),
+  ...priced(variantIds("claude-sonnet-5", QUALITY_EFFORTS), CLAUDE_SONNET_5),
+  ...priced(
+    [
+      "claude-sonnet-4-6",
+      "claude-sonnet-4-6-thinking",
+      "claude-sonnet-4-6-1m",
+      "claude-sonnet-4-6-thinking-1m",
+    ],
+    CLAUDE_SONNET_4_6
+  ),
+  MODEL_PRIVATE_11: CLAUDE_HAIKU_4_5,
+
+  ...priced(variantIds("gpt-5-6-sol", GPT_EFFORTS), GPT_5_6_SOL),
+  ...priced(priorityVariantIds("gpt-5-6-sol"), GPT_5_6_SOL_FAST),
+  ...priced(variantIds("gpt-5-6-terra", GPT_EFFORTS), GPT_5_6_TERRA),
+  ...priced(priorityVariantIds("gpt-5-6-terra"), GPT_5_6_TERRA_FAST),
+  ...priced(variantIds("gpt-5-6-luna", GPT_EFFORTS), GPT_5_6_LUNA),
+  ...priced(priorityVariantIds("gpt-5-6-luna"), GPT_5_6_LUNA_FAST),
+
+  ...priced(variantIds("kimi-k3", ["max", "high", "low"]), {
+    input: 3,
+    cached: 0.3,
+    output: 15,
+  }),
+  "kimi-k2-7": { input: 0.95, cached: 0.19, output: 4 },
+  ...priced(variantIds("glm-5-3", ["max", "high", "low"]), {
+    input: 1.4,
+    cached: 0.26,
+    output: 4.4,
+  }),
+  ...priced(variantIds("glm-5-3-flash", ["max", "high", "low"]), {
+    input: 0.15,
+    cached: 0.03,
+    output: 0.5,
+  }),
+
+  ...priced(["swe-1-7", "swe-1-7-medium"], {
+    input: 0.5,
+    cached: 0.2,
+    output: 2.5,
+  }),
+  ...priced(["swe-1-7-lightning", "swe-1-7-lightning-medium"], {
+    input: 2.5,
+    cached: 1,
+    output: 12.5,
+  }),
+  adaptive: { input: 0.5, cached: 0.1, output: 2 },
+  ...priced(variantIds("grok-4-6", ["xhigh", "high", "medium", "low"]), {
+    input: 2,
+    cached: 0.3,
+    output: 6,
+  }),
+  ...priced(variantIds("inkling", ["max", "xhigh", "high", "medium", "low", "none"]), {
+    input: 1.4,
+    cached: 0.26,
+    output: 4.4,
+  }),
+  ...priced(variantIds("deepseek-v4-flash", ["max", "high", "low"]), {
+    input: 0.14,
+    cached: 0.03,
+    output: 0.28,
+  }),
+  ...priced(variantIds("nemotron-3-ultra", ["high", "medium", "none"]), {
+    input: 0.6,
+    cached: 0.12,
+    output: 2.4,
+  }),
+  ...priced(variantIds("gemini-3-7-flash", ["high", "medium", "low"]), {
+    input: 1.5,
+    cached: 0.15,
+    output: 7.5,
+  }),
+  ...priced(variantIds("gemini-3-1-pro", ["high", "low"]), {
+    input: 2,
+    cached: 0.2,
+    output: 12,
+  }),
+  ...priced(variantIds("deepseek-v4-pro", ["max", "high", "low"]), {
+    input: 1.32,
+    cached: 0.04,
+    output: 3.96,
+  }),
+};
+
+// Each transport gets its own provider namespace. They share today's upstream
+// rate snapshot, but can diverge independently if Devin changes one channel.
+export const DEFAULT_PRICING_DEVIN = {
+  "devin-cli": { ...DEVIN_MODEL_PRICING },
+  dv: { ...DEVIN_MODEL_PRICING },
+  "devin-desktop": { ...DEVIN_MODEL_PRICING },
+  "devin-cli-agentic": { ...DEVIN_MODEL_PRICING },
+  dva: { ...DEVIN_MODEL_PRICING },
+};
--- a/src/shared/constants/pricing/frontier-labs.ts
+++ b/src/shared/constants/pricing/frontier-labs.ts
@@ -8,6 +8,7 @@ import {
  GPT_5_6_LUNA_PRICING,
  GPT_5_6_SOL_PRICING,
  GPT_5_6_TERRA_PRICING,
+  CLAUDE_FABLE_5_1_PRICING,
  CLAUDE_FABLE_5_PRICING,
  CLAUDE_OPUS_5_PRICING,
  CLAUDE_OPUS_4_PRICING,
@@ -213,6 +214,7 @@ export const DEFAULT_PRICING_FRONTIER = {
    // Common model IDs (without dates) used across providers
    // Intentional duplicates of dot-notation variants (e.g. claude-opus-4.6)
    // to cover hyphen-notation IDs (claude-opus-4-6) used by some clients
+    "claude-fable-5-1": CLAUDE_FABLE_5_1_PRICING,
    "claude-fable-5": CLAUDE_FABLE_5_PRICING,
    "claude-opus-5": CLAUDE_OPUS_5_PRICING,
    "claude-sonnet-5": CLAUDE_SONNET_5_PRICING,
--- a/src/shared/constants/pricing/oauth-subscriptions.ts
+++ b/src/shared/constants/pricing/oauth-subscriptions.ts
@@ -3,6 +3,7 @@
 * Pure data; merged by default-pricing.ts via spread (god-file decomposition; semantic split).
 */
 import {
+  CLAUDE_FABLE_5_1_PRICING,
  CLAUDE_OPUS_5_PRICING,
  GEMINI_3_7_FLASH_PROMO_PRICING,
  GPT_5_3_CODEX_PRICING,
@@ -20,6 +21,7 @@ const ANTIGRAVITY_GEMINI_3_7_PRICING = {

 export const DEFAULT_PRICING_OAUTH = {
  cc: {
+    "claude-fable-5-1": CLAUDE_FABLE_5_1_PRICING,
    "claude-fable-5": {
      input: 10.0,
      output: 50.0,
--- a/src/shared/constants/pricing/shared-tiers.ts
+++ b/src/shared/constants/pricing/shared-tiers.ts
@@ -60,6 +60,14 @@ export const CLAUDE_FABLE_5_PRICING = {
  cache_creation: 15.0,
 };

+export const CLAUDE_FABLE_5_1_PRICING = {
+  input: 10.0,
+  output: 50.0,
+  cached: 0.25,
+  reasoning: 50.0,
+  cache_creation: 12.5,
+};
+
 export const CLAUDE_OPUS_5_PRICING = {
  input: 5.0,
  output: 25.0,