Anthropic's Claude Fable 5.2 tops Artificial Analysis' AA-3D benchmark at 1271 Elo
In brief
- Claude Fable 5.2 ranked first on Artificial Analysis' AA-3D benchmark, scoring 1271 Elo.
- Other Claude models held second and third place on the AA-3D leaderboard.
- GPT-6 Astra trails by 103 points and Gemini 4 Argon trails by 124.
- Anthropic priced Fable 5.2 at $8/$40 per million input/output tokens, 20% below the earlier version, per Crypto Briefing.
How AA-3D scores models
AA-3D runs models through 55 tasks spanning several domains, including games and natural phenomena. Scores use Elo (yes, the rating system borrowed from chess).
When the benchmark launched, several frontier AI models were tied at the top. That's no longer the picture.
What does a 67-point lead actually mean?
Artificial Analysis lists Fable 5.2 at 1271 ±25, with a 95% confidence interval of 1246 to 1297, according to Crypto Briefing. That puts it 67 points clear of Sonnet 5.5, and even the bottom of the range (1246) sits above Sonnet 5.5's 1204.
It's not a rounding-error win.
The rivals sit further back. GPT-6 Astra trails Fable 5.2 by 103 points, and Gemini 4 Argon trails by 124, per the Artificial Analysis figures Crypto Briefing cited. The generational jump is notable too: Fable 5.2 beat its predecessor, Fable 5.1, by 132 points, and 5.1's score of 1139 had placed it below both GPT-6 Astra and Gemini 4 Argon. Inside Anthropic's own lineup, Sonnet 5.5 and Opus 5.5 sit just two points apart.
Pricing and safeguards
Anthropic has priced Fable 5.2 at $8 per million input tokens and $40 per million output tokens, which Crypto Briefing said is 20% cheaper than the earlier version. The same report said Fable 5.2 shares its underlying weights with Mythos 5.2, but includes stronger safeguards against misuse than Mythos 5.2 does.
Crypto Briefing also said the model maintained a leading position across various coding and scientific benchmarks, citing research it didn't identify. Those findings weren't detailed in the report, so the AA-3D numbers are the ones with a published score, error bar and confidence interval attached.
Frequently asked questions
What does the AA-3D benchmark measure?
AA-3D, from independent evaluator Artificial Analysis, measures how well AI models build 3D experiences. It runs models through 55 tasks across several domains, including games and natural phenomena, and scores them using Elo, the rating system borrowed from chess.
How far ahead is Claude Fable 5.2 on AA-3D?
Artificial Analysis lists Fable 5.2 at 1271 ±25 Elo, 67 points ahead of Sonnet 5.5's 1204, according to Crypto Briefing. GPT-6 Astra trails Fable 5.2 by 103 points and Gemini 4 Argon trails by 124.
How much does Claude Fable 5.2 cost?
Anthropic priced Fable 5.2 at $8 per million input tokens and $40 per million output tokens, according to Crypto Briefing. The report said that's 20% cheaper than the earlier version.
