whatmodel.sh / Claude Fable 5.2

Claude Fable 5.2 is a closed-weights language model from Anthropic, released November 2026. Licence: proprietary.

Catalogue identity notes

API model id claude-fable-5-2 (a dateless pinned snapshot). Released November 9, 2026. Anthropic: 'Claude Fable and Claude Mythos are the same underlying model'; Fable 5.2 is the general-availability configuration with cybersecurity and biology safeguards (blocked requests fall back to Claude Opus 4.8 for cyber and Claude Opus 5 for biology). Scores are Anthropic's own for the Fable 5.2 configuration unless a note says otherwise.

Model information sources

Release date ↗

88.5±0.8#1 of 743 LLMsfrom 21 benchmark results across 5 sources

Score within the LMArena (text) comparison set only.

$ how it compares

The LLMs ranked either side of Claude Fable 5.2, out of 897 with a score, then more from Anthropic.

#modelscorewhereRAMreleased
1Claude Fable 5.2 Anthropic89±1api $16.00/M2026-11
2GPT-6 Astra OpenAI84±1api $20.00/M2026-09
3Gemini 4 Carbon Google84±1api $4.00/M2026-10
4Claude Opus 5.5 Anthropic84±1api $8.00/M2026-09
5GPT-6.1 Sol OpenAI84±1api $4.00/M2026-09
more from Anthropic
7Claude Fable 5.1 Anthropic82±1api $20.00/M2026-09
8Claude Sonnet 5.5 Anthropic81±1api $4.00/M2026-09
11Claude Opus 5 Anthropic80±1api $10.00/M2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▄▅▅▃▃▂▃ #1 of 319 · 43% of its score · 8 tests
testresultvs the others in its rundifficultycountswhy
Arena Agent (October 2026 aggregate IPS)0.3035 IPS ✎▁▁▁▃▃█▂▁▁▁·▃ beats 63/63avg -0.403 IPS▇▇▇▇▇▇▇ 18%sharp
SWE-bench Pro (split unspecified) lab93.0% ✎▄█▄·▄██··▄·▄ beats 10/10very hard · avg 7.2%▇▇ 6.0%sharp · maxed out
Terminal-Bench 4.0 lab67.6% ✎▃·····▃··▅▅█ beats 7/8very hard · avg 0.7%▇▇ 5.2%sharp
Terminal-Bench 4.165.8% ✎▂▂▂··▅·▂▅█▃▆ beats 20/20very hard · avg 1.0%▇ 3.7%sharp · at its level · noisy
Humanity's Last Exam with tools lab74.2% ✎▂·····▂·▅█▂▃ beats 11/11very hard · avg 14%▇ 3.3%typical
SWE-bench Multilingual lab96.6% ✎▃·▃·▅▅█··▃·▃ beats 10/10hard · avg 31%▇ 3.0%maxed out
DeepSWE v1.1 (resolved) lab74.3% ✎▄·····▄▄▄▄·█ beats 6/6hard · avg 23%▇ 2.1%typical
Chartography with tools lab91.6% ✎▃·······▃▃██ beats 8/8very hard · avg 6.5%▇ 1.6%sharp · maxed out
coding▁▁▁▃▆▇█▆▆▃▃▃ #1 of 474 · 22% of its score · 4 tests
testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1837.8 ✎▁▂▄▃▇█▅▇▇▅▂▃ beats 134/134avg 996▇▇▇▇ 11%sharp
SWE-bench Pro (split unspecified) lab93.0% ✎▄█▄·▄██··▄·▄ beats 10/10very hard · avg 7.2%▇▇ 6.0%sharp · maxed out
SWE-bench Multilingual lab96.6% ✎▃·▃·▅▅█··▃·▃ beats 10/10hard · avg 31%▇ 3.0%maxed out
DeepSWE v1.1 (resolved) lab74.3% ✎▄·····▄▄▄▄·█ beats 6/6hard · avg 23%▇ 2.1%typical
math▁▁▃▅█▆▅█▄▁▁▃ #2 of 440 · 6.2% of its score · 3 tests

Trails GPT-6.1 Sol here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
FrontierMath tier 4 v2100.0% ✎▁▂▁▁█▅▂▃▁▁▁▃ beats 71/72very hard · avg 0.1%▇ 3.3%sharp · maxed out
FrontierMath tiers 1–3 v295.1% ✎▁··▁▁▂▅▃█▅▄▃ beats 91/91very hard · avg 1.4%▇ 2.8%sharp · maxed out
OTIS mock AIME 2024–202599.7% ✎▃▃▄▃▃▃▇▆█▄▃▆ beats 196/207very hard · avg 7.4%▇ 0.0%sharp · maxed out · noisy
web search▂▁▂▆▅█▄▃▃▁▁▁ #7 of 85 * · 3.3% of its score · 1 test

Trails GPT-4o (2024-08-06), GPT-4.1 mini, GPT-4.1 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Humanity's Last Exam with tools lab74.2% ✎▂·····▂·▅█▂▃ beats 11/11very hard · avg 14%▇ 3.3%typical
other tests9 not in a task · 40% of its score
testresultvs the others in its rundifficultycountswhy
EBR-bench (Epoch)90.5% ✎▃▂▅▆▃█▅▆▃▃▂▃ beats 30/30very hard · avg 0.3%▇▇▇ 8.6%sharp
Furniture Assembly (Epoch)90.0% ✎█▇██▁▇▁▃▃▄▁▃ beats 38/38very hard · avg 1.0%▇▇▇ 7.6%sharp
Mystery Game Puzzles (Epoch)90.0% ✎█·▂▁▆▆▆▆▅▂▂▃ beats 83/83very hard · avg 9.4%▇▇▇ 6.7%sharp · noisy
Humanity's Last Exam lab73.2% ✎▂····▂·▃▂█▂▃ beats 11/11very hard · avg 6.0%▇▇ 6.2%typical
Chess Puzzles (Epoch)79.0% ✎█▁▁▁▁▂▄▂▃▃▂▃ beats 157/157very hard · avg 5.6%▇▇ 4.9%sharp · noisy
Global-MMLU lab96.0% ✎▄▄··▄▄·█▄··▄ beats 7/7medium · avg 64%▇ 1.9%maxed out
Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite)78.7% ✎▂▂▄▂▃▃▂▇▇█▇▃ beats 47/47very hard · avg 14%▇ 1.9%typical
SimpleQA Verified94.0% ✎▁▂▂▂▅█▄▃▂··▃ beats 91/91very hard · avg 7.4%▇ 1.7%typical
1 more test
testresultvs the others in its rundifficultycountswhy
GPQA Diamond93.9% ✎▁▁▁▁▂▃▅▄▄▆█▃ beats 207/221hard · avg 39%▇ 0.6%sharp · maxed out

$ where to run it

Through an API

api · $16.00/M · $8 in / $40 out per 1M tokens · 1M context · platform.claude.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ. Same as Anthropic's own pricing page (https://platform.claude.com/docs/en/about-claude/pricing, row 'Claude Fable 5.2': $8 / MTok base input, $40 / MTok output), observed 2026-11-12. Optional fast mode is $16 / $80 per MTok.

$ how the score was worked out · Epoch AI — internal evaluatio… 36%, Anthropic system cards and mo… 29%, LMArena leaderboard dataset 29% and 2 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Epoch AI — internal evaluations9CC BY 4.0
Anthropic system cards and model announcements (lab-reported scores) · maker-reported8Factual vendor-reported score (not licensed content)
LMArena leaderboard dataset2CC BY 4.0
Terminal-Bench 4.0 official leaderboard submissions1Apache-2.0
Miðeind Icelandic LLM Leaderboard1CC BY 4.0
prior · pull toward the middle
weighted average= 88.5

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json