Claude Fable 5.2 Takes the Top Spot on the AI Benchmark That Matters Most

Claude Fable 5.2 Takes the Crown on November 9

Anthropic released Claude Fable 5.2 on November 9, and it immediately took first place on the benchmark most people in the industry actually watch. On Artificial Analysis's Intelligence Index, Claude Fable 5.2 scored 64.6, passing Gemini 4 Carbon at 60.0. It is the first time Anthropic has held the top spot on that index, and it did it while cutting the price: at $8 per million input tokens and $40 per million output tokens, Fable 5.2 costs 20% less than Fable 5.1. Better and cheaper is a rare combination in this market, and it is the reason this launch matters more than the usual round of model announcements.

The model is available now in the Claude apps, through the API, and on AWS, Google Cloud, and Azure. That wide availability matters. A model that only exists in a demo does not change anything. One that ships everywhere on day one forces every rival to answer it, and the answers tend to arrive fast.

Claude Fable 5.2 in Plain Numbers

The headline figures are simple enough to hold in your head. The Intelligence Index score is 64.6, the highest Artificial Analysis has recorded. Gemini 4 Carbon, announced October 28, sits at 60.0. The gap is not enormous, but the direction is what counts: Anthropic's previous flagship, Fable 5.1, scored 50.0 on the same index at its highest setting. A jump of nearly 15 points in one generation is the kind of move that resets what buyers expect from a frontier model.

Pricing tells the other half of the story. Fable 5.2 costs $8 in and $40 out per million tokens, with cached input at $0.20. That is 20% cheaper than Fable 5.1 while scoring far higher, which means the cost per unit of capability fell by a lot more than 20%. For teams running heavy workloads, that math lands directly on the monthly bill. Anthropic is clearly betting that volume beats margin here, and that a cheaper top model pulls developers off rival platforms faster than a pricey one ever could.

The Rest of November's Model Rush

Fable 5.2 did not arrive alone. November has already produced seven model launches, and the pace is the story as much as any single release. OpenAI shipped GPT-Image-2.6 on November 4, Tencent released Hy4 the same day, and Cognition put out the SWE-2 Flash coding model. November 5 brought Google's Gemini Omni 1.2 Flash video model and Thinking Machines Lab's Inkling 2 Small. Then November 9 delivered both Fable 5.2 and Meta's Muse Ember, the smallest model Meta has shipped, priced at $0.10 in and $0.50 out with a 36.0 index score.

Anthropic also launched Mythos 5.2 the same day, but that one is restricted to verified organizations, so most developers will never touch it. The pattern is worth noting: the frontier labs are now splitting their releases between models anyone can call and models that need approval. OpenAI did the same thing with GPT-6 Cyber, its security model that requires an application. The open frontier is still open, but the most sensitive capabilities increasingly sit behind a gate.

For anyone who wants a model they can actually download, Kimi K3.1 Pro remains the one to watch. Its weights went up on Hugging Face on November 2, it scores 46.5 on the index, and it is the most capable downloadable model available. Hy4 and Inkling 2 Small have both promised open weights without delivering them yet, and promises in this industry have a habit of slipping.

Why the Benchmark Lead Matters, and Why It Does Not

Benchmarks are a useful shorthand and a terrible final verdict. The Intelligence Index measures a basket of tasks, and a 64.6 says Fable 5.2 is the strongest general model on that basket today. It does not say it is the best model for your specific job. Coding, long document work, and agent-style tasks each have their own leaders, and the gaps there are often bigger than the headline number suggests.

What the lead does signal is momentum. Anthropic spent much of the year chasing Google and OpenAI on raw capability while winning on coding and agent workflows. Taking the top benchmark spot, at a lower price, changes the conversation from "Anthropic is good at agents" to "Anthropic has the best model, full stop." That is a different sales pitch, and it lands at a moment when enterprises are signing multi-year model deals.

The honest caveat is that leads in this market last weeks, not years. Gemini 4 Carbon held the top spot for less than two weeks. OpenAI has already teased its next GPT release, and the GPT-6.2 Astra leaks suggest something is coming soon. Fable 5.2 is the best model you can buy today, and today is doing a lot of work in that sentence.

What to Do With This Information

If you already use Claude, the upgrade is a straightforward win: better scores, lower price, same API. Test your heaviest prompts against it before switching everything over, because a model that wins on a benchmark can still surprise you on your own workload. If you are on another platform, the 20% price cut is the number to plug into your own cost model, not the 64.6. Benchmarks do not pay your cloud bill.

The broader lesson from November is that the model race has become a volume game. Seven launches in nine days, prices falling while scores rise, and the most capable downloadable model now comes from a Chinese lab that posted 2.8 trillion parameters of weights in public. The Inkling 2 Small launch showed the same pattern: hosted now, open weights promised next. The frontier is moving fast, and it is getting cheaper to stand on it. That is good news for everyone except the labs paying to build it.