Updated on September 30, 2026
Claude Sonnet 5.5 is Anthropic's mid-tier model, released on September 28, 2026, six days after the flagship Opus 5.5. On Terminal-Bench 4.0, an agentic coding test, it even edges past its bigger sibling with 70.6% against 66.4%. The API rate card stays identical to Sonnet 5, while output speed climbs by over 30%. Anthropic pitches it at well-scoped everyday work, from bug fixes to slide decks.
Claude Sonnet 5.5 generates output over 30% faster than Sonnet 5 and cuts per-task cost by up to 30%, according to Anthropic's launch figures. The saving comes from efficiency rather than a price cut, since the model burns fewer tokens and batches its tool calls, with a third fewer calls on some evaluations.
In practice, you paste a company's quarterly results and a slide template, and it builds the deck while double-checking numbers against the source documents. Anthropic aims it at clearly scoped jobs, fixing bugs, building spreadsheets, polishing interfaces. One telling detail, it is the first Sonnet to beat Pokémon Red working from screenshots alone.
Within the model family started in 2023), Sonnet sits in the middle, between the light Haiku and the top-tier Opus. A Claude Haiku 5.5 is slated for the coming weeks.
Claude Sonnet 5.5 scores 1,844 on GDPval-AA, a test built on real occupational tasks, against 1,846 for Opus 5.5, which costs twice as much per token. That reframes the old Sonnet versus Opus question. The mid-tier model now covers most scoped work, while Opus keeps the edge on ambiguous problems demanding sustained judgment.
Effort levels matter too. At Low or Medium, Sonnet 5.5 already tops Sonnet 5's best score for roughly a tenth of the per-task cost, again per Anthropic. That kind of price-performance shuffle is reshaping the ranking of language models, where OpenAI's GPT-6 Sol plays the same pricing game.
| Benchmark | Sonnet 5.5 | Opus 5.5 | Sonnet 5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 66.4% | 10.3% |
| OSWorld 2.1 | 80.1% | 81.8% | 57% |
| GDPval-AA | 1,844 | 1,846 | 1,449 |
| CursorBench 4.0 | 55.5% | 57.8% | 34.1% |
A million tokens costs $2 on input and $10 on output, with cache reads at $0.20 and cache writes at $2.50. Opus 5.5 charges double, $4 and $20. On the app side, Sonnet 5.5 replaces Sonnet 5 as Claude's mid-tier model, under the terms of each plan.
Developers reach it on the Anthropic API under the claude-sonnet-5-5 identifier, plus Amazon Bedrock, Google Cloud Vertex AI and Microsoft Azure (a relief for cautious IT departments). Those figures reflect launch day, and Anthropic's published price list remains the only reference that ages well.
Sonnet 5.5 beats Opus 5.5 on Terminal-Bench 4.0 with 70.6% against 66.4%, at half the price per token. Opus 5.5 still leads on CursorBench 4.0 (57.8% versus 55.5%) and on large, ambiguous refactors requiring sustained judgment. For everyday coding and debugging, Sonnet 5.5 is the smarter starting point.
Anthropic's lineup stacks three tiers, Haiku as the lightest, Sonnet in the middle and Opus at the top, alongside the gated Fable models. Sonnet 5.5 costs $2 and $10 per million tokens, Opus 5.5 exactly twice that. On GDPval-AA, the performance gap has narrowed to two points, 1,844 versus 1,846.
Inside the Claude apps, Sonnet 5.5 replaces Sonnet 5 as the mid-tier model, with access and usage limits that depend on your plan, worth confirming on claude.ai. Through the API, it runs at $2 per million input tokens and $10 for output, exactly Sonnet 5's rate.
Over 30% faster on output speed, per Anthropic, making it the quickest Sonnet released so far. Per-task cost drops by up to 30% because the model consumes fewer tokens and tool calls. Benchmarks jump as well, from 10.3% to 70.6% on Terminal-Bench 4.0 and from 57% to 80.1% on OSWorld 2.1.
Verdict: For scoped coding, spreadsheets and slide decks, this is the model to try first, since the upgrade costs nothing extra. Teams already living in Claude, from solo developers to enterprise squads on Bedrock or Azure, get near-Opus results on a Sonnet budget.
Claude Opus 5.5 
GPT-6 Astra 
Claude Fable 5.1 
