Anthropic Claude Opus 5.5 Cuts Output Price, Beats Fable in Benchmarks
Anthropic launches Claude Opus 5.5 at $20 per million output tokens, claiming Fable-beating results while applying strict biology and cyber safeguards.
Summary
Anthropic released Claude Opus 5.5 on Tuesday, September 22, 2026, claiming state of the art coding and knowledge work performance. Its most capable Claude tier reportedly beats the larger Fable on many benchmarks and completed informal tasks Fable failed. Output costs $20 per million tokens, down from Opus 5’s $25, with similar reductions on other metrics, faster serving and lower compute. Opus 5.5 also uses less jargon and puts important information first. Opus sits above midrange Sonnet and Haiku, the fastest and cheapest tier.
Arriving two months after Opus 5’s July 24 launch, it precedes Sonnet 5.5 and Haiku 5.5, promised in coming weeks with similar improvements. Anthropic says Opus 5.5 is comparable to Mythos in biology and cybersecurity, triggering Fable safeguards that restrict exploit discovery in compiled programs and development of recognizable biological weapons. It is Anthropic’s first model since CEO Dario Amodei backed pacing frontier capabilities so alignment can catch up. Safety training remained broadly similar to predecessors, including alignment testing and prerelease evaluations by METR and Frontier Design. Anthropic is preparing stronger training, evaluation, security and monitoring for future models, wants public policy to assume a larger safety role, and plans to share more details soon.
Positives
- Output pricing falls from $25 to $20 per million tokens compared with Opus 5.
- Opus 5.5 reportedly surpasses the larger Fable across many benchmarks and completes informal tasks Fable could not.
- Lower serving compute makes Opus 5.5 faster to run alongside its price reductions.
- Less jargon and earlier placement of important information should make responses clearer.
- Sonnet 5.5 and Haiku 5.5 are due in coming weeks with similar performance improvements.
Risks & concerns
- Biology and cybersecurity capabilities comparable to Mythos require safeguards against exploit discovery and recognizable biological weapons development.
- Safety training remained broadly similar to previous models despite Opus 5.5’s increased capabilities.
- Stronger training, evaluation, security and monitoring systems are still being prepared for future models.
- Anthropic says public policy capacity needs time to develop as increasingly capable systems create larger safety demands.