Anthropic released its new artificial intelligence model Opus 5.5 on 22 September 2026. The company announced that the model had set a new performance standard in coding and knowledge work, surpassed the larger Fable in many tests, and completed some unofficial tasks that Fable could not complete.
The three tiers of the Claude series are as follows:
- Opus: The most capable and most expensive tier
- Sonnet: The mid-tier
- Haiku: The fastest and least expensive tier
Opus 5.5 was said to operate faster than the previous version and require less computing power to run. The price of output tokens was reduced to $20 per million tokens, compared with $25 for the previous model. The model’s communication style also changed: its likelihood of using jargon was reduced, while its tendency to present important information at the beginning of messages was increased.
The release came just two months after Opus 5, which was released on 24 July. Anthropic said that the subsequent tiers, Sonnet 5.5 and Haiku 5.5, would be released “in the coming weeks” and would offer similar performance improvements.
The company stated that Opus 5.5 was on par with Mythos in biology and cybersecurity, and was therefore subject to the safety measures applied to Fable. The measures limit uses such as discovering vulnerabilities in compiled programs and developing recognizable biological weapons.
The model was the first release since CEO Dario Amodei called for progress in AI capabilities to be matched to the pace of safety work. While its safety training was similar to that of previous models, METR and Frontier Design conducted external evaluations. Anthropic announced that it was preparing more advanced safety and monitoring systems for future models.
Background
Anthropic is not a new name in the FikirPilot archive: we have published 31 articles mentioning this name in the past 90 days; the latest was dated October 3, 2026.