🤖 feat: promote Claude Opus 5.1 as the opus model

.github/workflows/auto-cleanup-fixup.yml

@@ -118,6 +118,6 @@ jobs:
           FAILED_RUN_ID: ${{ github.event.workflow_run.id }}
         run: |
           bunx mux@next run \
-            --model anthropic:claude-opus-5 \
+            --model anthropic:claude-opus-5-1 \
             --thinking xhigh \
             < .github/prompts/auto-cleanup-fixup.md

.github/workflows/auto-cleanup.yml

@@ -71,6 +71,6 @@ jobs:
           GH_TOKEN: ${{ steps.app-token.outputs.token }}
         run: |
           bunx @coder/xum@next run \
-            --model anthropic:claude-opus-5 \
+            --model anthropic:claude-opus-5-1 \
             --thinking xhigh \
             < .github/prompts/auto-cleanup.md

.github/workflows/nightly-terminal-bench.yml

@@ -10,7 +10,7 @@ on:
   workflow_dispatch:
     inputs:
       models:
-        description: 'Models to test (comma-separated, or "all" for opus-5 + gpt-5.6-sol + google/gemini-3-pro-preview + google/gemini-3-flash-preview + google/gemini-3.7-flash)'
+        description: 'Models to test (comma-separated, or "all" for opus-5-1 + gpt-5.6-sol + google/gemini-3-pro-preview + google/gemini-3-flash-preview + google/gemini-3.7-flash)'
         required: false
         default: "all"
         type: string
@@ -129,7 +129,7 @@ jobs:
           INPUT_MODELS: ${{ inputs.models }}
         run: |
           if [ "$INPUT_MODELS" = "all" ] || [ -z "$INPUT_MODELS" ]; then
-            echo 'models=["anthropic/claude-opus-5","openai/gpt-5.6-sol","google/gemini-3-pro-preview","google/gemini-3-flash-preview","google/gemini-3.7-flash"]' >> "$GITHUB_OUTPUT"
+            echo 'models=["anthropic/claude-opus-5-1","openai/gpt-5.6-sol","google/gemini-3-pro-preview","google/gemini-3-flash-preview","google/gemini-3.7-flash"]' >> "$GITHUB_OUTPUT"
           else
             # Convert comma-separated to JSON array
             models_json=$(echo "$INPUT_MODELS" | jq -R -s -c 'split(",") | map(gsub("^\\s+|\\s+$"; ""))')

.github/workflows/terminal-bench.yml

@@ -4,7 +4,7 @@ on:
   workflow_call:
     inputs:
       model_name:
-        description: "Model to use (e.g., anthropic/claude-opus-5, openai/gpt-5.6-sol)"
+        description: "Model to use (e.g., anthropic/claude-opus-5-1, openai/gpt-5.6-sol)"
         required: false
         type: string
       dataset:
@@ -98,7 +98,7 @@ on:
         required: false
         type: string
       model_name:
-        description: "Model to use (e.g., anthropic/claude-opus-5, openai/gpt-5.6-sol)"
+        description: "Model to use (e.g., anthropic/claude-opus-5-1, openai/gpt-5.6-sol)"
         required: false
         type: string
       mux_run_args:

benchmarks/terminal_bench/prepare_leaderboard_submission.py

@@ -123,6 +123,13 @@
         "model_org_display_name": "Anthropic",
         "folder_name": "Claude-Opus-5",
     },
+    "anthropic/claude-opus-5-1": {
+        "model_name": "claude-opus-5-1",
+        "model_provider": "anthropic",
+        "model_display_name": "Claude Opus 5.1",
+        "model_org_display_name": "Anthropic",
+        "folder_name": "Claude-Opus-5.1",
+    },
     # Keep historical GPT metadata alongside the current GPT-5.6 Sol bench target
     # so mixed or older artifact sets still map to the canonical leaderboard names.
     "openai/gpt-5.2": {

docs/config/models.mdx

@@ -15,7 +15,7 @@ Xum ships with curated models kept up to date with the frontier. Use any custom
 | ---------------------- | ----------------------------- | ------------------------------------------------------------ | ------- |
 | Fable 5.1              | anthropic:claude-fable-5-1    | `fable`                                                      |         |
 | Mythos 5               | anthropic:claude-mythos-5     | `mythos`                                                     |         |
-| Opus 5                 | anthropic:claude-opus-5       | `opus`                                                       | ✓       |
+| Opus 5.1               | anthropic:claude-opus-5-1     | `opus`                                                       | ✓       |
 | Sonnet 5               | anthropic:claude-sonnet-5     | `sonnet`                                                     |         |
 | Haiku 4.5              | anthropic:claude-haiku-4-5    | `haiku`                                                      |         |
 | GPT-5.6 Sol            | openai:gpt-5.6-sol            | `gpt`, `sol`                                                 |         |

docs/guides/github-actions.mdx

@@ -116,7 +116,7 @@ jobs:
           GH_TOKEN: ${{ steps.app-token.outputs.token }}
         run: |
           bunx @coder/xum@next run \
-            --model anthropic:claude-opus-5 \
+            --model anthropic:claude-opus-5-1 \
             --thinking xhigh \
             < .github/prompts/auto-cleanup.md
 ```

src/browser/features/ChatInput/useCreationWorkspace.test.tsx

@@ -1360,7 +1360,7 @@ describe("useCreationWorkspace", () => {
     });
     expect(onWorkspaceCreated.mock.calls[0][1]).toEqual({
       autoNavigate: true,
-      pendingStreamModel: "anthropic:claude-opus-5",
+      pendingStreamModel: "anthropic:claude-opus-5-1",
       markPendingInitialSend: false,
     });
   });
@@ -1838,7 +1838,7 @@ describe("useCreationWorkspace", () => {
     expect(onWorkspaceCreated.mock.calls.length).toBe(1);
     expect(onWorkspaceCreated.mock.calls[0][1]).toEqual({
       autoNavigate: true,
-      pendingStreamModel: "anthropic:claude-opus-5",
+      pendingStreamModel: "anthropic:claude-opus-5-1",
       markPendingInitialSend: true,
     });
   });

src/cli/run.test.ts

@@ -171,7 +171,7 @@ describe("xum CLI", () => {
     test("shows default model as opus", async () => {
       const result = await runCli(["run", "--help"]);
       expect(result.exitCode).toBe(0);
-      expect(result.stdout).toContain("anthropic:claude-opus-5");
+      expect(result.stdout).toContain("anthropic:claude-opus-5-1");
     });
 
     test("--service-tier has no default auto", async () => {

src/common/config/schemas/appConfigOnDisk.ts

@@ -123,6 +123,13 @@ export const AppConfigMigrationsSchema = z
      * source key.
      */
     defaultModelFallbacksSeededFable51: z.boolean().optional(),
+    /**
+     * One-time chain-target migration after the opus alias moved to Opus 5.1:
+     * chains still exactly equal to the superseded shipped default
+     * (Fable 5.1 → Opus 5) move to the current default (Fable 5.1 → Opus 5.1);
+     * any user-edited chain is left untouched.
+     */
+    defaultModelFallbacksSeededOpus51: z.boolean().optional(),
     /** One-time migration from the legacy auto-delete default to persistent sub-agents. */
     persistentSubagentsDefaulted: z.boolean().optional(),
   })

src/common/constants/knownModels.test.ts

@@ -3,7 +3,11 @@
  */
 
 import { describe, test, expect } from "@jest/globals";
-import { KNOWN_MODELS, MODEL_ABBREVIATIONS } from "@/common/constants/knownModels";
+import {
+  KNOWN_MODELS,
+  MODEL_ABBREVIATIONS,
+  TOKENIZER_MODEL_OVERRIDES,
+} from "@/common/constants/knownModels";
 import modelsJson from "@/common/utils/tokens/models.json";
 import { findMissingKnownModels } from "@/common/utils/tokens/updateModelsData";
 
@@ -19,6 +23,16 @@ describe("Known Models Integration", () => {
     }
   });
 
+  test("opus alias tracks Opus 5.1 and retired ids keep tokenizer overrides", () => {
+    expect(MODEL_ABBREVIATIONS.opus).toBe("anthropic:claude-opus-5-1");
+    expect(KNOWN_MODELS.OPUS.id).toBe("anthropic:claude-opus-5-1");
+    // Exact-id lookup for retired-but-documented custom model strings must keep
+    // resolving to the Opus 4.5 approximation instead of falling back (with a
+    // warning) to the generic per-provider tokenizer.
+    expect(TOKENIZER_MODEL_OVERRIDES["anthropic:claude-opus-5"]).toBe("anthropic/claude-opus-4.5");
+    expect(TOKENIZER_MODEL_OVERRIDES[KNOWN_MODELS.OPUS.id]).toBe("anthropic/claude-opus-4.5");
+  });
+
   test("gemini-flash resolves to the stable Gemini 3.7 Flash model", () => {
     expect(MODEL_ABBREVIATIONS["gemini-flash"]).toBe("google:gemini-3.7-flash");
   });

src/common/constants/knownModels.ts

@@ -56,15 +56,15 @@ const MODEL_DEFINITIONS = {
     // approximate counting; real usage can run ~1.0-1.3x higher.
     tokenizerOverride: "anthropic/claude-opus-4.5",
   },
-  // Claude Opus 5 - released July 24, 2026. Successor to Opus 4.8 at the same pricing
-  // ($5/M input, $25/M output). API id `claude-opus-5`; Opus 4.8 stays usable as the
-  // custom model string `anthropic:claude-opus-4-8`.
+  // Claude Opus 5.1 - successor to Opus 5 (released July 24, 2026) at unchanged pricing
+  // ($5/M input, $25/M output). API id `claude-opus-5-1`; Opus 5 stays usable as the
+  // custom model string `anthropic:claude-opus-5`.
   OPUS: {
     provider: "anthropic",
-    providerModelId: "claude-opus-5",
+    providerModelId: "claude-opus-5-1",
     aliases: ["opus"],
     warm: true,
-    // Opus 5 uses the newer Opus 4.7+ tokenizer (~30% more tokens for the same text),
+    // Opus 5.1 uses the newer Opus 4.7+ tokenizer (~30% more tokens for the same text),
     // which isn't published upstream; reuse Opus 4.5 for approximate counting. Real
     // usage can run ~1.0-1.3x higher than this estimate (same situation as FABLE above).
     tokenizerOverride: "anthropic/claude-opus-4.5",
@@ -280,6 +280,7 @@ export const MODEL_ABBREVIATIONS: Record<string, string> = Object.fromEntries(
 // lookup does not fall back to the generic per-provider tokenizer.
 const LEGACY_TOKENIZER_MODEL_OVERRIDES: Record<string, string> = {
   "anthropic:claude-fable-5": "anthropic/claude-opus-4.5",
+  "anthropic:claude-opus-5": "anthropic/claude-opus-4.5",
   "anthropic:claude-opus-4-8": "anthropic/claude-opus-4.5",
 };
 

src/common/utils/ai/modelDisplay.test.ts

@@ -10,6 +10,7 @@ describe("formatCompactModelDisplayName", () => {
 
   test("falls back to the full display name when no shorter label is clearer", () => {
     expect(formatCompactModelDisplayName("claude-opus-5")).toBe("Opus 5");
+    expect(formatCompactModelDisplayName("claude-opus-5-1")).toBe("Opus 5.1");
     expect(formatCompactModelDisplayName("gpt-5.3-codex-spark")).toBe("Spark 5.3");
   });
 });

src/common/utils/ai/modelFallbacks.ts

@@ -17,7 +17,9 @@ export const MODEL_FALLBACK_CHAIN_LIMIT = 3;
  *
  * Seeded into the config exactly once, guarded by
  * migrations.defaultModelFallbacksSeeded (plus the one-shot
- * defaultModelFallbacksSeededFable51 re-seed for the key move to Fable 5.1)
+ * defaultModelFallbacksSeededFable51 re-seed for the key move to Fable 5.1,
+ * and the one-shot defaultModelFallbacksSeededOpus51 target migration for the
+ * Opus 5.1 promotion, see SUPERSEDED_DEFAULT_MODEL_FALLBACKS)
  * — on versions that know the flag,
  * user edits or deletions of these chains are never overridden by updates.
  * (Versions predating the flag strip it on save, so a downgrade→save→
@@ -37,14 +39,32 @@ export const DEFAULT_MODEL_FALLBACKS: ModelFallbacks = {
  * refusal fallback intact.
  */
 export const LEGACY_DEFAULT_MODEL_FALLBACKS: ModelFallbacks = {
-  "anthropic:claude-fable-5": { models: [KNOWN_MODELS.OPUS.id] },
+  // Target pinned to the literal Opus 5 id (not KNOWN_MODELS.OPUS.id): this
+  // chain exists for downgraded pre-5.1 builds, whose own shipped default was
+  // Fable 5 → Opus 5. Pinning keeps the seeded bytes identical no matter which
+  // build ran the original seed pass.
+  "anthropic:claude-fable-5": { models: ["anthropic:claude-opus-5"] },
+};
+
+/**
+ * Shipped defaults superseded by the Opus 5.1 promotion, keyed like
+ * DEFAULT_MODEL_FALLBACKS. Guarded by migrations.defaultModelFallbacksSeededOpus51,
+ * config load rewrites a chain to the current default exactly once IF the stored
+ * entry still deep-equals the superseded default (same single-model chain, no
+ * enabled/triggers customization). Any deviation is user intent and is never
+ * touched; the legacy `anthropic:claude-fable-5` chain is deliberately not
+ * migrated so downgraded builds keep their expected bytes on disk.
+ */
+export const SUPERSEDED_DEFAULT_MODEL_FALLBACKS: ModelFallbacks = {
+  [KNOWN_MODELS.FABLE.id]: { models: ["anthropic:claude-opus-5"] },
 };
 // Deep-freeze: entries are spread by reference into live configs (fresh-install
 // defaults, seed merge). Accidental in-place mutation must crash fast instead
 // of silently corrupting the process-wide default.
 for (const entry of [
   ...Object.values(Object.freeze(DEFAULT_MODEL_FALLBACKS)),
   ...Object.values(Object.freeze(LEGACY_DEFAULT_MODEL_FALLBACKS)),
+  ...Object.values(Object.freeze(SUPERSEDED_DEFAULT_MODEL_FALLBACKS)),
 ]) {
   Object.freeze(entry);
   Object.freeze(entry.models);

src/common/utils/ai/models.test.ts

@@ -143,9 +143,15 @@ describe("Anthropic 1M context classification", () => {
     expect(hasNative1MContext("anthropic:claude-sonnet-4-5")).toBe(false);
   });
 
-  it("treats Opus 4.6 to 4.8, Opus 5, and Sonnet 4.6 as native 1M models", () => {
+  it("treats Opus 4.6 to 4.8, Opus 5, Opus 5.1, and Sonnet 4.6 as native 1M models", () => {
     expect(getAnthropic1MContextMode("anthropic:claude-opus-5")).toBe("native");
     expect(getAnthropic1MContextMode("anthropic:claude-opus-5-20260724")).toBe("native");
+    // The claude-opus-5 pattern only tolerates date suffixes, so the 5.1 id
+    // needs (and has) its own pattern.
+    expect(getAnthropic1MContextMode("anthropic:claude-opus-5-1")).toBe("native");
+    expect(getAnthropic1MContextMode("mux-gateway:anthropic/claude-opus-5-1")).toBe("native");
+    expect(hasNative1MContext("anthropic:claude-opus-5-1")).toBe(true);
+    expect(supports1MContext("anthropic:claude-opus-5-1")).toBe(false);
     expect(getAnthropic1MContextMode("anthropic:claude-opus-4-8")).toBe("native");
     expect(getAnthropic1MContextMode("anthropic:claude-opus-4-8-20260528")).toBe("native");
     expect(getAnthropic1MContextMode("anthropic:claude-opus-4-6")).toBe("native");

src/common/utils/ai/models.ts

@@ -182,6 +182,7 @@ const ANTHROPIC_NATIVE_1M_PATTERNS = [
   new RegExp(`^claude-fable-5-1${OPTIONAL_VERSION_SUFFIX}$`, "i"),
   new RegExp(`^claude-mythos-5${OPTIONAL_VERSION_SUFFIX}$`, "i"),
   new RegExp(`^claude-opus-5${OPTIONAL_VERSION_SUFFIX}$`, "i"),
+  new RegExp(`^claude-opus-5-1${OPTIONAL_VERSION_SUFFIX}$`, "i"),
   new RegExp(`^claude-opus-4-8${OPTIONAL_VERSION_SUFFIX}$`, "i"),
   new RegExp(`^claude-opus-4-7${OPTIONAL_VERSION_SUFFIX}$`, "i"),
   new RegExp(`^claude-opus-4-6${OPTIONAL_VERSION_SUFFIX}$`, "i"),

src/common/utils/ai/providerOptions.test.ts

@@ -139,7 +139,12 @@ describe("buildProviderOptions - Anthropic", () => {
   // Native-xhigh models (Opus 4.7+ / Sonnet 5+): xhigh is a distinct native
   // effort and adaptive thinking requires `display: "summarized"` to return
   // thinking content.
-  for (const model of ["claude-opus-4-7", "claude-opus-5", "claude-sonnet-5"] as const) {
+  for (const model of [
+    "claude-opus-4-7",
+    "claude-opus-5",
+    "claude-opus-5-1",
+    "claude-sonnet-5",
+  ] as const) {
     describe(`${model} (native xhigh effort + summarized display)`, () => {
       for (const { thinking, expectedThinking, effort } of [
         {

src/common/utils/thinking/policy.test.ts

@@ -423,6 +423,16 @@ describe("getThinkingPolicyForModel", () => {
       "xhigh",
       "max",
     ]);
+    // Opus 5.1 rides the same Opus 5+ wildcard: full 6-level ladder, and
+    // (unlike Mythos-class models) "off" stays available.
+    expect(getThinkingPolicyForModel("anthropic:claude-opus-5-1")).toEqual([
+      "off",
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+      "max",
+    ]);
     expect(getThinkingPolicyForModel("anthropic:claude-opus-5-0")).toEqual([
       "off",
       "low",

src/common/utils/tokens/models-extra.ts

@@ -194,6 +194,25 @@ export const modelsExtra: Record<string, ModelData> = {
     supports_response_schema: true,
   },
 
+  // Claude Opus 5.1 - successor to Opus 5 at the same pricing/shape: $5/M input,
+  // $25/M output, cache write 1.25x input / cache read 0.1x input, native 1M
+  // context, 128K max output, full effort ladder with native xhigh and max.
+  "claude-opus-5-1": {
+    max_input_tokens: 1000000,
+    max_output_tokens: 128000,
+    input_cost_per_token: 0.000005, // $5 per million input tokens
+    output_cost_per_token: 0.000025, // $25 per million output tokens
+    cache_creation_input_token_cost: 0.00000625, // $6.25 per million tokens (1.25× input)
+    cache_read_input_token_cost: 0.0000005, // $0.50 per million tokens (0.1× input)
+    litellm_provider: "anthropic",
+    mode: "chat",
+    supports_function_calling: true,
+    supports_vision: true,
+    supports_pdf_input: true,
+    supports_reasoning: true,
+    supports_response_schema: true,
+  },
+
   // Claude Opus 5 - Released July 24, 2026
   // Same pricing/shape as Opus 4.8: $5/M input, $25/M output, native 1M context
   // (both default and maximum), 128K max output, full effort ladder with native

src/common/utils/tools/tools.test.ts

@@ -73,6 +73,7 @@ describe("supportsAnthropicNativeWebFetch", () => {
     ["claude-mythos-5", true],
     // Two-segment IDs at/after the 4.6 cutoff.
     ["claude-fable-5-1", true],
+    ["claude-opus-5-1", true],
     ["claude-sonnet-4-6", true],
     ["claude-opus-4-6", true],
     ["claude-opus-4-8", true],