What changed
Anthropic released Claude Opus 5.5 on September 22, 2026 as the first model in the Claude 5.5 family. It is priced at $4 per million input tokens and $20 per million output tokens, 20% below Opus 5 list pricing. Anthropic says the new model performs at the level of Claude Fable 5.1 for most tasks while costing about 40% less per task than Opus 5 because it also uses fewer steps and tokens. The model launched across Anthropic's own products and API plus AWS, Google Cloud, Microsoft Azure and GitHub Copilot.
Why it matters
Frontier-model economics are increasingly determined by completed-task cost rather than token price alone. If Opus 5.5 can deliver Fable-class coding and knowledge-work capability with lower list prices and fewer inference steps, builders can move workloads down from the most expensive tier without accepting the capability drop that usually accompanies that choice. That also puts pressure on routing systems and fixed model tiers: the economically best model may change even when an application's task mix does not.
Anthropic is compressing its own model ladder
Anthropic positions Opus 5.5 as comparable with the larger Fable 5.1 on most work rather than merely as an incremental Opus update. Reuters reports list pricing of $4/M input and $20/M output, 20% below Opus 5, while Anthropic's task-level claim is a roughly 40% cost reduction because the model also needs less compute and fewer steps.
Early independent benchmarks support the efficiency story
Independent benchmark coverage from Artificial Analysis places Opus 5.5 at or near the top of current intelligence and cost-performance comparisons, including strong coding and knowledge-work results. Benchmark rankings remain workload-specific, but they provide a useful check beyond Anthropic's own launch tables.
Distribution is immediate rather than staged
The model launched through major cloud providers as well as Anthropic's own API, and GitHub Copilot added it on launch day for agentic coding and long-running tasks. That makes the economics actionable immediately for teams already using those distribution layers rather than requiring a provider migration.
The safety boundary is also changing
Anthropic says external groups including METR and Frontier Design tested the model before launch. The company reports stronger containment behavior and applies safeguards associated with its highest-capability systems. Some sensitive cyber and biology requests may be routed through older models under those controls, so the effective model behind a request can depend on safety classification.