Anthropic's Sonnet 5.5 comes within two points of Opus 5.5 on computer use at unchanged $2 and $10 per million tokens

Anthropic shipped Claude Sonnet 5.5 on September 28. It describes the model as more than 30% faster than Sonnet 5 and up to 30% cheaper for most work. The list price did not change: $2 per million input tokens and $10 per million output tokens. The saving comes from completing tasks with fewer tokens.

On Anthropic's own launch table, Sonnet 5.5 nearly matches Opus 5.5. It scores 80.1% against Opus 5.5's 81.8% on OSWorld 2.1, a computer-use benchmark, and 1,844 against 1,846 on GDPval-AA v2.1 for knowledge work. It beats Opus 5.5 on Terminal-Bench 4.0, scoring 70.6% against 66.4%. These are vendor-reported results on Anthropic's own test harness, and no third party has reproduced them yet.

Opus 5.5 arrived on September 22. Anthropic said it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. Third-party comparisons put Opus 5.5's list prices 60% below Fable 5.1 and found that it outscores Fable 5.1 on every benchmark in the launch table. When a mid-tier model reaches the flagship's level on the vendor's own evals within three weeks of the flagship's release, either the flagship is priced for a narrow set of hard tasks or the evals no longer separate the tiers. The launch materials do not settle which.

Competitors' prices keep falling below Anthropic's. Meta priced Muse Spark 1.1 at $1.25 input and $4.25 output per million tokens in July, and Mark Zuckerberg described that as roughly a quarter of what Anthropic and OpenAI charge for comparable models. Anthropic has responded through token efficiency rather than list-price cuts. That protects its headline rate card, while effective cost per task falls for customers who measure it.