Claude’s most advanced models for coding and knowledge work. Their research capabilities also offer an early glimpse of how AI models will contribute to scientific progress.
the pricing changes are what actually matter day to day rather than being a marketing footnote - cheaper cache reads and lower cost for highly agentic runs mean I stopped rationing long tool-use sessions the way I used to. it also showed up same-day in the gateway I already use (Netlify), so switching over didn't mean touching any existing keys or config.
What needs improvement
the launch mentions cybersecurity false positives dropping 60%, but doesn't say whether that's from genuinely better judgment or a loosened threshold - a published tradeoff number (false negatives, not just false positives) would make that claim easier to trust rather than take on faith.
GPT-4o is fast and flexible for general chat and quick tasks, but Fable 5.1 is specifically tuned for long agentic runs and coding - the lower effort modes matching the previous version's output quality while costing less made it the easier default for the multi-step work I do most.
GPT-4o