Anthropic has released Claude Opus 5.5, the first model in its 5.5 family, with lower prices than Opus 5 and performance it says approaches Claude Fable 5.1 on most work. The model is available through Anthropic’s platform and on Amazon Web Services, Google Cloud and Microsoft Azure.
The pricing change has two parts. Input tokens cost $4 per million and output tokens $20 per million, each 20 percent below Opus 5. Cache reads fall from $0.50 to $0.20 per million tokens. Anthropic says the new model also uses fewer tokens to finish typical jobs, producing a 40 percent reduction in total workload cost at its default settings. That larger figure is a company estimate for tasks, not the cut in the standard per-token rate.
The company reports improvements in coding and computer-use evaluations. On Terminal-Bench 4.0, a test of multi-step command-line work, Anthropic lists a score of 66.4 percent for Opus 5.5, against 52.3 percent for Opus 5 and 55.8 percent for Fable 5.1. Those figures come from Anthropic’s release and its stated test setup; they are not a single independent ranking of every model under identical conditions. Anthropic also cautions that benchmark margins can exaggerate the gap between Opus 5.5 and Fable in everyday use.
For developers running coding agents, the combination of task cost and completion rate matters more than price per token alone. An agent that takes fewer steps can reduce the bill even before a price cut. Anthropic says Opus 5.5 generates output more than 30 percent faster than Opus 5 in its tests, and offers a separate fast mode at higher rates.
The launch follows Anthropic’s decision to offer Fable 5.1 and Mythos 5.1 with different safeguards. Anthropic says Opus 5.5 has biology and cybersecurity capabilities comparable to Mythos 5.1, so it is applying controls similar to Fable’s. Most cybersecurity tasks are redirected to an older Opus model; ordinary bug-finding and repair in software development remain available. Vetted life-sciences organizations can apply for broader access now, while expanded verification for cybersecurity practitioners is planned.
Anthropic says outside evaluators, including METR and Frontier Design, tested the model before release. Its published results on alignment and prompt-injection resistance are company claims, and the company acknowledges limits in assessing how models behave outside tests. The stronger safeguards could also affect some benchmark runs, because restricted tasks may be handled by another model.
The lower standard rates make Opus 5.5 a cheaper option for long-running agent work than its predecessor. Whether its reported gains translate into the same savings for a given customer will depend on the work, effort setting, caching and safeguard routing used.
Sources: Anthropic
–
By the Control Plane Editorial Team