Opus 5.5 Brings Flagship-Level AI Down in Price
Anthropic released Opus 5.5 on September 22, 2026. The company called it the strongest-performing model it had tested to date. It said Opus 5.5 beats its larger Fable model on many benchmarks, the standard tests used to compare AI models. Output is priced at $20 per million tokens, the small chunks of text a model reads and writes, down from $25 for the previous model. TechCrunch
The gap from the prior release was about two months. Opus 5 arrived on July 24. Anthropic said Sonnet 5.5 and Haiku 5.5 will be released in the coming weeks. The Opus revision follows the September 1 introduction of Claude Fable 5.1 and Claude Mythos 5.1, which Anthropic describes as its most advanced models for coding and knowledge work. Company materials describe Opus 5.5 as performing at the level of Claude Fable 5.1 on most work.
Anthropic said Opus 5.5 is comparable to Mythos for biology and cybersecurity capabilities. It placed Opus 5.5 under the same safeguards as its Fable model. Those safeguards restrict use for finding security flaws in compiled programs, software shared without readable source code, and for developing recognizable biological weapons. Opus 5.5 was checked before release by outside organizations including METR and Frontier Design. TechCrunch
Company materials describe Opus 5.5 as costing 40% less to run than Opus 5, in addition to the $20 per million output token price. Anthropic describes Claude Opus 5 as offering greatly improved performance for the same cost as its predecessor, Opus 4.8. Pricing for Claude Opus 4.8 for regular use is $5 per million input tokens and $25 per million output tokens. Pricing for Claude Opus 4.7 remains the same as Opus 4.6, at $5 per million input tokens and $25 per million output tokens.
Anthropic describes Claude Opus 5 as a step-change improvement for the Opus tier, with stronger coding, more capable agents that can carry out multi-step tasks, and sharper professional work. It also describes Opus 5 as thoughtful and proactive, and as coming close to the frontier intelligence of Claude Fable 5. Earlier in the series, Anthropic updated Claude Opus 4.6 in February 2026 to carry out financial research, an update that came days after the company's push into legal services. Anthropic has said its priority was to safely release Fable as soon as it could, even at the cost of overly broad safeguards.
On September 19, Reuters reported that Anthropic was considering releasing a new AI model ahead of its IPO, its first sale of stock to the public. Reuters also reported in September 2026 that Anthropic could push its IPO to after the November U.S. midterm elections. Reuters
The broader context here is useful for teams that run models in production. A smaller model that matches a flagship on many tests, while cutting the cost of running it to answer queries, changes capacity planning. It lowers the cost penalty for sending demanding coding, agent, and knowledge work to Opus instead of saving it for Fable. It also puts more weight on testing and guardrails, since biology and cybersecurity abilities now sit in a cheaper and more widely used tier with Fable-grade controls.
In my view, the pace is worth watching. Opus 5 in July, Fable 5.1 and Mythos 5.1 on September 1, Opus 5.5 on September 22, with Sonnet 5.5 and Haiku 5.5 still to come. That pace shortens the time buyers have to set prompts, testing tools, and cost controls before the next revision arrives. I have watched this cycle before with PCs and cloud, and the hopeful pattern is familiar. When capable tools get cheaper to run, more experiments fit the budget, and more of them grow into lasting tools.
