# Claude Fable 5.2 sets record on Epoch Capabilities Index at 174 > Source: > Published: 2026-11-09 21:26:12+00:00 # Claude Fable 5.2 sets record on Epoch Capabilities Index at 174 Photo: Tima Miroshnichenko / Pexels Anthropic's newest model edges past OpenAI's GPT-6 Astra on two major leaderboards and arrives with a lower price than its predecessor [Anthropic](https://cryptobriefing.com/markets/anthropic/) has a new top scorer. Claude Fable 5.2 posted a 174 on the Epoch Capabilities Index, a new record that clears [OpenAI](https://cryptobriefing.com/markets/openai/)’s GPT-6 Astra and its score of 169. The model launched on November 9, 2026. Anthropic built it for complex reasoning and agentic tasks, meaning work where the AI plans and carries out multistep jobs instead of answering one question and stopping. ## The scoreboard, line by line The Epoch result is the headline number, but it is not the only leaderboard Fable 5.2 now sits atop. On the Artificial Analysis Intelligence Index, the model scored 65, the highest mark on that ranking. [Google](https://cryptobriefing.com/markets/alphabet/)’s Gemini 4 Carbon trails at 60. Anthropic’s own Claude Opus 5.5 lands at 59, six points back. ### AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. The biggest jumps showed up in specialized, harder tests. Fable 5.2 hit 80% on HLE-Diamond and landed between 73% and 75% on Terminal-Bench-Science. ## Cheaper, but still not cheap Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. That pricing marks a 20% reduction from Fable 5.1. For the uninitiated, tokens are the small chunks of text a model reads and writes. Input is what you send in, output is what the model generates, and here output costs five times as much as input. Even with the discount, Fable 5.2 remains a premium product. It still sits above Claude Opus 5.5 on price, so Anthropic is effectively running a tiered menu inside its own lineup. The technical specs back up the premium positioning. The model supports a 1 million-token context window and can produce up to 128,000 tokens of output in a single response. The model is also described as performing robustly in high-effort settings, where it is given more room to work through a problem before responding. **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/).