Anthropic · Model guide
Claude Fable 5.2
Anthropic's most capable generally available model and the successor to Claude Fable 5.1, built for demanding reasoning and long-horizon agentic work: agent sessions that run for hours, multistep deep research, and analysis carried through to a finished document, spreadsheet or deck. Thinking is always on and effort (default high) is the only depth control; forced tool choice is rejected, and editing earlier turns invalidates its thinking blocks on every account. Claude Mythos 5.2 is the same model with different safeguards, available only to organizations verified through Anthropic's verification programs.
What this model is good at
What the vendor positions it for, and which of our models to reach for instead.
VendorFor demanding reasoning and long-horizon agentic work.source
Agentic coding · Anthropic reports 49.1% on FrontierCode 2.0 at xhigh effort, the highest of any released model it tested, and first place on Proximal's FrontierSWE v2 ultra-long-horizon engineering tasks.(vendor claim)source
Multistep research · Anthropic positions it for multistep deep research and reports 74.6% on Terminal-Bench-Science 0.1 against 52.6% for Claude Fable 5.1.(vendor claim)source
Professional knowledge work · Anthropic reports a 41.3% automation rate on the Remote Labor Index, about double Claude Opus 5.5 and Claude Fable 5.1, and the highest rubric score on AA-Briefcase v1.1 in its comparison table.(vendor claim)source
- anthropic/claude-opus-5-5cheaper —Anthropic recommends starting with the latest Claude Opus model for most workloads, and moving to Claude Fable 5.2 only when evals on Opus at higher effort still fall short.(vendor claim)source
- anthropic/claude-fable-5-1predecessor —Claude Fable 5.1 still accepts the earlier computer_20251124 computer use tool, which returns a 400 on this model outside Amazon Bedrock.source
Pricing and billing
Billed per token. Cached input is charged at the cache-read rate.
| Price / 1M tokens | You pay |
|---|---|
| Input | $ 8.00 |
| Output | $ 40.00 |
| Cache read | $ 0.20 |
| Cache write | $ 10.00 |
+ 2.0K × $ 40.00
Capabilities and limits
What CrossModel guarantees across every route this model can take right now.
- Context window
- 1.0M tokens
- Max output
- 128.0K tokens
- Input / output modalities
- Text + Image → Text
- Streaming
- Supported
- Tool calling
- Supported
- Structured output (JSON)
- Supported
- Image input
- Supported
- Thinking mode
- Supported· always on
- Reasoning effort
- low · medium · high · xhigh · max
- Available endpoints
- /v1/chat/completions · /v1/responses · /v1/messages
Published benchmarks
Scores the vendor reported, with the evaluation setup each one came from.
- FrontierCode 2.0score·Anthropic · 2026-11-09effort xhighClaude Codesetup not fully disclosed
Run and reported by Cognition in Claude Code; mean of five runs per task. Rebuilt 132-task suite with looser scope grading, not comparable with FrontierCode v1.1. Each model's best effort level: xhigh here; Anthropic reports 45.3% at high and 45.7% at max (44.4% if the 22 timed-out max-effort attempts are scored zero). Same table: Claude Opus 5.5 45.5% (max), Claude Fable 5.1 32.4% (xhigh), GPT-6 Astra 42.8%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
49.1% - FrontierSWE v2score·Anthropic · 2026-11-09effort maxProximalsetup not fully disclosed
Run by Proximal in its own agent harness: 34 ultra-long-horizon engineering and research tasks, 20-hour budget per trial, mean of five trials per task. Same evaluation: GPT-6 Astra 65.5%, Claude Opus 5.5 62.3%, Claude Sonnet 5.5 61.9%, Claude Fable 5.1 56.3%.
69.9% - Terminal-Bench 4.0pass_rate·Anthropic · 2026-11-09effort maxClaude Codesetup not fully disclosed
Claude Code in --bare mode, 10 trials per task (660 trials), no internet access (needed resources pre-cached; offline web search and fetch over a frozen index on most tasks); standard error ±1.8 pts. With safeguards on, 2.5% of requests (10% of trials) were answered by the fallback model. Same section: Claude Opus 5.5 66.4% (at xhigh), Claude Fable 5.1 55.8%.
67.6%
- AutomationBenchscore·Anthropic · 2026-11-09effort maxsetup not fully disclosed
Run by Zapier on its private held-out AutomationBench 1.0.6 set (657 tasks), with the API's default fallbacks; not the 600-task public set behind this catalog's AutomationBench v1.0.6 rows. 39.7% at low effort. Same leaderboard: Claude Opus 5.5 42.5% at max (its stored 40.0% is Zapier's earlier run without fallbacks), Claude Fable 5.1 31.4%, GPT-6 Astra 41.4%.
51.4%
- SWE-Bench Proresolved_rate·Anthropic · 2026-11-09effort maxsetup not fully disclosed
From the system card's capability summary (Table 8.1.A); average over five trials. Same table: Claude Opus 5.5 89.9%, Claude Fable 5.1 81.2%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
93%
- OSWorld 2.1 (offline subset, partial score)score·Anthropic · 2026-11-09effort maxsetup not fully disclosed
The benchmark's official offline subset: 82 of 108 tasks, VM without internet access. Mean per-task checkpoint credit, Pass@1 averaged over five attempts, 1080p, up to 500 action steps; strict pass rate 54.6%. Re-evaluated under the same conditions: Claude Opus 5.5 87.2%, Claude Fable 5.1 83.5%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
89.3%
- AA-Briefcase v1.1score·Anthropic · 2026-11-09effort maxsetup not fully disclosed
Artificial Analysis long-horizon knowledge-work Elo rating, not a percentage; run independently by Artificial Analysis. Rubric pass rate 66.1%, the highest in the table. At xhigh effort Anthropic reports 1738. Same table: Claude Opus 5.5 1800, Claude Fable 5.1 1669, GPT-6 Astra 1569. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
1785
- Chartography (no tools)score·Anthropic · 2026-11-09effort maxsetup not fully disclosed
Five runs, graded by Gemini 3.5 Flash following the Surge AI leaderboard. With tools Anthropic reports 91.6%. Same table: Claude Opus 5.5 64.4%, Claude Fable 5.1 44.8%, GPT-6 Astra 71.0%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
73.4%
- Humanity's Last Examaccuracy·Anthropic · 2026-11-09effort maxsetup not fully disclosed
No tools, 980k-token task budget; graded by Claude Opus 4.6. From the system card's capability summary (Table 8.1.A); same table: Claude Opus 5.5 64.4%, Claude Fable 5.1 60.9%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
73.2% - Humanity's Last Exam (with tools)accuracy·Anthropic · 2026-11-09effort maxsetup not fully disclosed
Tools were web search, web fetch, programmatic tool calling and code execution, with a 980k-token task budget and HLE-discussing sources blocklisted; graded by Claude Opus 4.6. Same table: Claude Opus 5.5 67.7%, Claude Fable 5.1 65.6%, GPT-6 Astra 57.2%. Anthropic reports Claude Fable 5.2 with production safeguards on; flagged requests were answered by a fallback model under the default server-side fallback policy.
74.2%
Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.
Use it in your tools
Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.
Frequently asked questions
What is Claude Fable 5.2?
How much does Claude Fable 5.2 cost?
Does Claude Fable 5.2 support tool calling and structured output?
Which endpoint do I call?
Can I try it without writing code?
Chat first, integrate later
Chat first, then wire it in once you like the answers.