Claude Fable 5.2 takes first place on an independent AI ranking, ahead of Gemini 4 Carbon

What the ranking says

As Ifanr reported, Anthropic's new Claude Fable 5.2 has moved to the top of a closely watched third-party leaderboard. The ranking comes from Artificial Analysis, an independent testing firm, and the list is called Intelligence Index v5. Fable 5.2 scored 65 points.

That puts it 5 points ahead of Gemini 4 Carbon and 6 points ahead of Claude Opus 5.5, the previous Anthropic flagship. Fable 5.2 also came first in 5 of the 9 individual tests that feed the index. Its strongest areas were scientific reasoning, general knowledge and programming.

The headline framing, that it beat Opus 5.5 overnight, fits with this. Anthropic released Fable 5.2 and a sibling model, Mythos 5.2, on the same day, and the independent ranking followed.

Anthropic's own benchmark numbers

Anthropic published its own results too. On FrontierCode 2.0, a test of how well an AI agent can write code, Fable 5.2 scored 49.1%. Opus 5.5 got 45.5%, GPT-6 Astra got 42.8% and Gemini 4 Carbon got 41.1%.

On FrontierSWE v2, a software engineering test, Fable 5.2 reached 69.9%. That is 13.6 percentage points above the previous generation. These are the company's figures, so treat them as a claim until others reproduce them. The Artificial Analysis result matters because an outside party produced it.

The price of first place

The top score comes with a bill. At its strongest reasoning setting, Fable 5.2 averages $26.5 per task on the index, about 2.6 times what Opus 5.5 costs for the same kind of work.

Anthropic's numbers show a way around this. On FrontierCode 2.0, Fable 5.2 at a lower reasoning effort came close to Opus 5.5 at its highest effort, at roughly half the cost per task. In other words, the cheaper setting of the new model can match the old model's best.

Standard API pricing is $8 per million input tokens and $40 per million output tokens, about 20% below the previous generation. A token is a small chunk of text, roughly a word fragment, and API pricing counts them. A Fast Mode generates text up to 2.5 times quicker but doubles the rates to $16 and $80.

Why Anthropic itself says not to switch everything

Anthropic positions the Fable line for hard engineering, math and research work. For everyday coding and general knowledge tasks, the company still recommends Claude Opus 5.5. That is an unusual thing for a vendor to say on launch day, and it supports the view that the cheapest strong model is often the right choice.

Early user reaction is mixed in an informative way. AI entrepreneur Matt Shumer called it the best model he had used and said it handled projects that other models could not. He also said it can be overly cautious, sometimes blocking ordinary work tasks.

There is another limit. Anthropic says it deliberately held back the model's gains in areas that could speed up building the next generation of AI, such as optimizing code for AI chips and improving language model design. In those tasks it stays at roughly Opus 5.5 level, so a high overall rank does not mean it improved at everything.

How to read a leaderboard

An index like this blends several tests into one number. That is handy for building a shortlist, but it averages across tasks that may look nothing like your week. A model that wins 5 of 9 tests still loses 4.

It is also a snapshot. Anthropic released these models a little over two months after the last pair, and the Fable 5.2 lead over Gemini 4 Carbon is 5 points. Rankings can change quickly when any lab ships an update.

Access is another factor. Fable 5.2 is available through Claude and the AWS, Google Cloud and Microsoft Azure cloud platforms. Mythos 5.2, which keeps fuller capabilities, goes mainly to vetted organizations. From November 9 to 16, eligible Claude Pro and standard Team users can use Fable models within 50% of their normal weekly allowance at no extra charge.