Anthropic has released Opus 5.5, a model that pushes performance higher while lowering the cost of every output token. Released on September 22, 2026, the model sets a new state-of-the-art in coding and knowledge work, runs faster, and changes how it communicates with users.
That combination gives Opus 5.5 a clear advantage: more capability, less compute required to serve it, and a lower price than the previous model. The release also places Anthropic’s latest system in direct comparison with Fable and Mythos across demanding areas of work.
A Faster Model With a Lower Price
Anthropic will charge $20 per million output tokens for Opus 5.5, down from $25 for the previous model. That price cut arrives alongside a faster run time, reflecting an overall drop in the compute required to serve the model.
The change matters because output tokens drive the cost of extended model use. A lower price can make long coding sessions and knowledge work less expensive, while the faster performance can reduce the time users spend waiting for results.
Opus 5.5 also set a new state-of-the-art in coding and knowledge work performance. Across many benchmarks, it outpaces the larger Fable model, showing that a bigger model does not always deliver the strongest result.
The gap is not limited to formal benchmark scores. Opus 5.5 succeeded in a number of informal tasks that Fable failed to complete, adding another layer to the comparison between the two models.
Communication Changes Alongside Capability
Anthropic made significant changes to how Opus 5.5 communicates. The model is less likely to use jargon and more likely to put important information at the start of its messages, giving users a clearer path to the answer.
Those changes reach beyond tone. When a model leads with the information that matters most, its responses become easier to scan and use, especially during coding and knowledge work. Less jargon can also make complex material more accessible.
Opus 5.5 is comparable to Mythos in its biology and cybersecurity capabilities. That comparison places the new model alongside Mythos in two areas that demand strong technical performance, even as Opus 5.5 claims a new state-of-the-art across coding and knowledge work.
The release presents a model with several distinct strengths:
- Opus 5.5 costs $20 per million output tokens, compared with $25 for the previous model.
- The model runs faster because serving it requires less compute.
- It sets a new state-of-the-art in coding and knowledge work performance.
- It outpaces the larger Fable model in many benchmarks.
- It completed informal tasks that Fable failed to complete.
- It uses less jargon and places important information earlier in messages.
Safety, Pacing, and the Next Frontier
Opus 5.5’s release is subject to safeguards similar to those of the Fable model. Its safety training was similar to that of its predecessors, with alignment testing and pre-release evaluation by METR and Frontier Design.
Anthropic CEO Dario Amodei has embraced pacing the frontier and slowing AI capabilities progress. He described the reason for that approach directly: “I have become convinced that fully addressing the risks requires even more prudence, not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”
That position shapes the wider release. Anthropic is preparing more advanced training and evaluation systems for future models, creating new systems to test capabilities and safety as its models develop.
The company has also started to put infrastructure in place to support public policy efforts on AI safety. The move connects model development with the broader work of creating safeguards and policy responses as AI capabilities advance.
Opus 5.5 arrives as a sharper, faster, and less expensive model, but its importance reaches beyond a single benchmark or price change. It combines stronger coding and knowledge work performance with communication changes, comparable biology and cybersecurity capabilities to Mythos, and safety measures tied to Anthropic’s effort to pace progress.
The next stage will test how far those advantages extend. Anthropic is preparing more advanced evaluation and training systems, while Opus 5.5 begins its run with a lower price, faster serving, and performance that has already moved beyond earlier comparisons.
Based on
