Claude Fable 5.2 takes first place on Artificial Analysis’ AA-3D benchmark

Anthropic's Claude Fable 5.2 debuted at 1271 Elo on Artificial Analysis' AA-3D benchmark, first place and 67 points ahead of the next model, Claude Sonnet 5.5 at 1204 Elo, according to the independent evaluator. Anthropic released Fable 5.2 on November 9, 2026, at $8 per million input tokens and $40 per million output tokens, 20% cheaper than predecessor Claude Fable 5.1, which scored 1139 Elo — a 132-point generational gain. The benchmark runs 55 tasks covering games and natural phenomena, and Artificial Analysis lists Fable 5.2's score with a ±25 margin of error and a 95% confidence interval of 1246 to 1297.

Claude Fable 5.2 takes first place on Artificial Analysis’ AA-3D benchmark Anthropic's newest model scored 1271 Elo on the 3D experience test, a 67-point lead over its own siblings, and costs 20% less than its predecessor Anthropic https://cryptobriefing.com/markets/anthropic/ has a new leader on one of AI’s more visual leaderboards. Claude Fable 5.2 ranks first on the AA-3D benchmark from independent evaluator Artificial Analysis, with a score of 1271 Elo . The company’s closest competition on the board is, awkwardly, itself. The next two spots belong to other Claude models. Anthropic unveiled Fable 5.2 on November 9, 2026. It is pitched as a model built for complex technical and agentic tasks, meaning work where the AI carries out multi-step jobs rather than answering one question at a time. The scoreboard AA-3D measures how well AI models build 3D experiences. The benchmark runs models through 55 tasks spanning several domains, including games and natural phenomena. Scores use Elo, the rating system borrowed from chess. A higher number means a model performs better relative to the others in the pool. Here is how the top of the table shakes out, according to Artificial Analysis: - Claude Fable 5.2: 1271 Elo - Claude Sonnet 5.5: 1204 Elo - Claude Opus 5.5: 1202 Elo - GPT-6 Astra: 1168 Elo - Gemini 4 Argon: 1147 Elo - Claude Fable 5.1: 1139 Elo That puts Fable 5.2 67 points clear of Sonnet 5.5. In a ranking where second and third place are separated by just two points, a 67-point gap is a fairly roomy lead. The score also carries a margin of error. Artificial Analysis lists it as 1271 ±25, with a 95% confidence interval of 1246 to 1297. Put simply, the evaluator is confident the true score lands somewhere in that band. Even the bottom of that range, 1246, sits above Sonnet 5.5’s 1204. A big jump from the last version The more striking number may be the generational improvement. Fable 5.2 beat its predecessor, Fable 5.1, by 132 points. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Fable 5.1 scored 1139, which placed it below both GPT-6 Astra and Gemini 4 Argon. One version later, the Fable line has gone from trailing rivals from OpenAI https://cryptobriefing.com/markets/openai/ and Google https://cryptobriefing.com/markets/alphabet/ to sitting above everyone. Fable 5.2’s lead extends beyond 3D work. The model also maintained a leading position across various coding and scientific benchmarks, according to the research findings. Cheaper, too Better performance usually costs more. Not this time. Fable 5.2 is priced at $8 per million input tokens and $40 per million output tokens. That is 20% cheaper than the earlier version. For the uninitiated, tokens are the chunks of text a model reads and writes. Input tokens cover what you send in, output tokens cover what comes back, and output is typically the pricier side of the bill. Same brain, different guardrails Fable 5.2 has a sibling. It shares its underlying weights with Mythos 5.2. Weights are the learned parameters that make a model what it is. Two models sharing weights are essentially the same engine under different hoods. The difference lies in the safety layer. Fable 5.2 includes stronger safeguards against misuse than Mythos 5.2 does. How we got here The top of AA-3D was not always this tidy. When the benchmark launched, several frontier AI models were tied at the top. Fable 5.2’s showing breaks that stalemate, at least for now. A clean lead on a leaderboard that previously had no single winner is a meaningful change in the standings. Artificial Analysis acts as a third-party referee here. Independent scoring matters in a field where every company has a marketing department eager to declare victory. What this means For Anthropic, the result is a strong talking point in a fight for developer attention. Holding the top three slots on a benchmark means rivals have to get past several Claude models, not just one. For OpenAI and Google, the gap is the story. GPT-6 Astra trails Fable 5.2 by 103 points, and Gemini 4 Argon trails by 124. There is also an internal wrinkle for Anthropic. Sonnet 5.5 and Opus 5.5 sit just two points apart on AA-3D, which suggests customers focused on this type of task may not see much difference between them here. Fable 5.2 now gives them a clearer default option. A few caveats are worth keeping in mind. AA-3D is one benchmark covering 55 tasks, and strong scores on a test do not automatically translate into strong results in every real-world project. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .