Peter Gostev pits Fable 5.2 against Opus across more than 100 generations

Anthropic’s newest model is getting the stress-test treatment.

In a video comparison, Peter Gostev pitted Claude Fable 5.2 against Opus across more than 100 generations and several creative tasks. According to the research summary of his evaluations, Gostev’s testing highlighted the Fable models as more insightful than competitors.

What Fable 5.2 brings to the table

Anthropic unveiled Claude Fable 5.2 on November 9, 2026, alongside a more advanced sibling, Claude Mythos 5.2.

The pitch is that both models are meaningful upgrades over prior versions. Anthropic is aiming them at demanding work in coding, reasoning, vision, and complex long-term decision-making.

Initial benchmarks indicate the new models beat their predecessors. They reportedly deliver results up to 30% faster, with improved clarity in their outputs.

The most striking number comes from the Remote Labor Index. Fable 5.2 scored 41.3% on that benchmark, while Opus 5.5 came in at 21.3%.

Running more than 100 generations helps smooth out the luck factor. Any model can produce one brilliant answer or one dud, so volume is what separates a pattern from a fluke.

Pricing, tiers, and who gets access

Anthropic set standard pricing at $8 per million input tokens and $40 per million output tokens.

There is also a fast mode for users who want speed over savings. That option doubles the rates to $16 per million input tokens and $80 per million output tokens.

Access is split by model. Extra usage of Fable is available to eligible paid subscribers.

Mythos is more gated. Access to the higher-end model is tied to Anthropic’s Life Sciences and Cyber Verification Programs.

How the Fable line got here

The Fable and Mythos family is still young. Anthropic introduced the line in June 2026 with the launch of Claude Fable 5.

Since then, follow-up releases have concentrated on three themes: speed, clarity, and specialized applications. Version 5.2 continues that trajectory rather than reinventing it.

What this means for Anthropic and its rivals

There are caveats. The 30% speed improvement is an upper bound, described as up to 30%, so real-world gains will vary by task.

The Remote Labor Index result is also a single benchmark. One strong score does not guarantee the same margin across every workload a business might throw at the model.

The gated rollout of Mythos is the other thread to follow. How Anthropic expands access beyond its Life Sciences and Cyber Verification Programs will signal how confident the company is in deploying its most capable model more broadly.