Claude Fable 5.2 is Anthropic's AI model, released 09/11/2026. Fable 5.2 is the generally available configuration of a new Anthropic model released alongside Claude Mythos 5.2, which has more permissive cyber and biology safeguards and is restricted to vetted organisations. The system card reports gains over Fable 5.1 and Opus 5.5 in coding, knowledge work and scientific reasoning. Unless noted, scores use adaptive thinking at max effort with safeguards on, averaged over five trials.
Visit Claude Fable 5.2AI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Claude Fable 5.2 score on independent benchmarks?
Claude Fable 5.2 ranks #1 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 64.6, as measured on 22 November 2026.
Where does Claude Fable 5.2 rank against other models?
How good is Claude Fable 5.2 at agentic coding?
How does Claude Fable 5.2 do on individual tests?
| Evaluation (run by Artificial Analysis) | Claude Fable 5.2 | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 59.6% | 67.2% |
| FrontierCode 2.0 | 48.8% | 48.8% |
| Humanity's Last Exam | 74.3% | 74.3% |
| GDPval-AA (Elo) | 1831 | 1866 |
How has Claude Fable 5.2's score changed over time?
We have tracked Claude Fable 5.2 on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Claude Fable 5.2 system card report?
Fable 5.2 is the generally available configuration of a new Anthropic model released alongside Claude Mythos 5.2, which has more permissive cyber and biology safeguards and is restricted to vetted organisations. The system card reports gains over Fable 5.1 and Opus 5.5 in coding, knowledge work and scientific reasoning. Unless noted, scores use adaptive thinking at max effort with safeguards on, averaged over five trials.
These are Anthropic's own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Sources: Claude Fable 5.2 and Claude Mythos 5.2 System Card (09/11/2026), Claude Fable 5.2 and Claude Mythos 5.2 (09/11/2026). Last checked 22/11/2026.
Coding
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| FrontierCode 2.0xhigh effort, each model at its best effort | 49.1% | 32.4% | 45.5% | 42.8% |
| FrontierCode 2.0 at max effortmax effort | 45.7% | n/a | 45.5% | n/a |
| FrontierSWE v2run by Proximal | 69.9% | 56.3% | 62.3% | 65.5% |
| SWE-bench Promax effort, mean of 5 trials | 93% | 81.2% | 89.9% | n/a |
| SWE-bench Multilingualmax effort, mean of 5 trials | 96.6% | 89.1% | 93.9% | n/a |
| SWE-bench Multimodalmax effort, mean of 5 trials | 74.1% | 54.7% | 61.4% | n/a |
| DeepSWE v1.1max effort, mean of 5 trials | 74.3% | n/a | n/a | n/a |
| Terminal-Bench 4.0safeguards on, no internet access | 67.6% | 55.8% | 66.4% | n/a |
Agents & tool use
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1safeguards on (Mythos 5.2 without safeguards: 75.3%) | 74.6% | 52.6% | 58.7% | 68.1% |
| AutomationBenchrun by Zapier | 51.4% | 31.4% | 42.5% | 41.4% |
| WANDR (relaxed score), agentic searchmax effort | 82.1% | 68.7% | 72.3% | n/a |
| OSWorld 2.1 (offline set, partial score)max effort; strict pass rate 54.6% | 89.3% | 83.5% | 87.2% | 79.5% |
| RoboDojo Simhighest available effort, run internally | 64.9% | 27.6% | 39.4% | 29.8% |
Reasoning & maths
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Humanity's Last Exam (no tools)max effort | 73.2% | 60.9% | 64.4% | n/a |
| Humanity's Last Exam (with tools)max effort | 74.2% | 65.6% | 67.7% | 57.2% |
Knowledge & science
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| GMMLU (42 languages)max effort, no tools | 96% | 94% | 94.3% | n/a |
| MILU (11 languages)max effort, no tools | 94.1% | 93% | 93.1% | n/a |
Multimodal
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Chartography (no tools)max effort | 73.4% | 44.8% | 64.4% | 71% |
| Chartography (with tools)max effort | 91.6% | 88.4% | 89% | n/a |
| BenchCAD 2.1max effort | 83.5 score (0 to 100) | 44.4 | 65.3 | n/a |
Other
| Benchmark | Claude Fable 5.2 | Claude Fable 5.1 | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Remote Labor Index (RLI)High effort, default settings | 41.3% | 17.9% | 21.3% | 20.8% |
| GDPval-AA v2.1max effort | 1831 Elo | 1763 | 1866 | 1584 |
| AA-Briefcase v1.1 (rubric)max effort | 66.1% | 61.5% | 61.2% | 52% |
| AA-Briefcase v1.1 (Elo)max effort | 1785 Elo | 1669 | 1800 | 1569 |
| CyScenarioBench (Claude Mythos 5.2, no safeguards)Mythos 5.2 configuration; each lab's score is based on unique tasks | 76.3% | n/a | n/a | n/a |
| CVE Rediscovery (Claude Mythos 5.2, no safeguards)Mythos 5.2 configuration | 89.6% | n/a | n/a | n/a |
| De novo protein binder design, 24 hr (Claude Mythos 5.2, no safeguards)Mythos 5.2 configuration | 85.1 score | n/a | n/a | n/a |
What did safety testing find for Claude Fable 5.2?
Safety level: Responsible Scaling Policy and Frontier Compliance Framework: CB-1 protections applied, CB-2 not crossed; AI R&D threshold not met; cyber in the lower category of the FCF
- Claude Mythos 5.2 does not cross the CB-2 threshold and CB-1 protections are applied; the AI R&D threshold is judged not met, with no 2x acceleration of Anthropic's aggregate AI R&D progress (an uncertain assessment).
- Mythos 5.2 has the strongest cyber capabilities of any Anthropic model released, yet still falls in the lower capability category of the FCF; Fable 5.2's cyber safeguards enforce the same policy as Opus 5.5 and no critical-severity or universal jailbreak was found.
- Fable 5.2 falls back to Opus 4.8 more often than recent models in prompt injection scenarios, giving a higher attack success rate, but when it answers an attempt itself it is one of the most robust models to date.
- Single-turn harmless response rate was lower than Fable 5.1 and multi-turn tracking and surveillance regressed (addressed with new classifiers); multi-turn biological weapons conversations improved.
- White-box assessments found Mythos 5.2 internally aware it may be in an evaluation in as many as 43% of automated behavioural audit transcripts, so results may not fully predict deployment behaviour; it attempted to cross a sandbox boundary less often than any other model tested.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our Claude Fable 5.2 deep dive.
Anthropic Model Timeline
Up to 1M tokens (system card: context window sizes are evaluation-dependent and do not exceed 1M tokens) context
Up to 1M tokens (system card: ProgramBench episodes run up to the full 1M token window) context
1M tokens context
1M tokens context
1M tokens context
1M tokens context
Long-context context
200k tokens context
200k tokens context
200k tokens context
1,000k tokens context
200k tokens context
200k tokens context
1,000k tokens context
200k tokens context
200k tokens context
200k tokens context
200k tokens context
200k tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
