Anthropic releases Opus 5.5 with lower prices and stronger benchmark performance than Fable
Anthropic says its new flagship outperforms the larger Fable model across many benchmarks, setting state-of-the-art results in coding and knowledge work. Output pricing falls 20%, from $25 to $20 per million tokens, while reduced compute makes it faster.
Anthropic released Opus 5.5 on Tuesday, September 22, 2026, describing it as “the strongest-performing model we’ve tested to date.” The company says the new model sets a state-of-the-art in coding and knowledge work, and outperforms the larger Fable model across many benchmarks.
Opus 5.5 is the top tier in Anthropic’s three-tier Claude lineup, which also includes Sonnet in the middle and Haiku as the fastest and cheapest option. The release comes just two months after Opus 5 launched on July 24, 2026. Anthropic says Sonnet 5.5 and Haiku 5.5 will follow “in the coming weeks” with similar performance improvements.
The new model is cheaper and faster than its predecessor. Output tokens are priced at $20 per million tokens, down from $25 for the previous model, with comparable reductions across other pricing metrics. The speed improvement reflects a reduction in the compute required to run it.
Anthropic also changed how Opus 5.5 communicates — the model is less likely to use jargon and more likely to lead with the most important information. The company says Opus 5.5’s biology and cybersecurity capabilities are comparable to its Mythos model, placing it under the same safeguards applied to Fable. Those safeguards restrict uses such as discovering exploits in compiled programs or developing biological weapons.
The release is the first since CEO Dario Amodei publicly endorsed slowing the pace of AI capabilities development. “I have become convinced that fully addressing the risks requires even more prudence,” Amodei wrote earlier this month, “not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.” Safety training for Opus 5.5 followed a similar approach to previous models, including alignment testing and pre-release evaluation by outside organizations such as METR and Frontier Design.
Anthropic noted that more advanced training and evaluation systems, including improved security and monitoring tools, are being prepared for future models. The company also indicated it is building infrastructure to support a larger role for public policy in AI safety oversight.