Claude Sonnet 5.5 ships as the faster everyday model

Anthropic introduced Claude Sonnet 5.5 on September 28, 2026, and the claim it leads with is not a new list price. The model is the second in the Claude 5.5 family, priced the same as Sonnet 5. Anthropic says it is a clear upgrade on that predecessor: it runs 30%+ faster and costs up to 30% less for most work. The official @claudeai post the same day used the same framing — more than 30% faster, and up to 30% less for most work.

Sonnet 5.5 is positioned as a faster, lower-cost complement to Claude Opus 5.5. Anthropic reserves Opus 5.5 for complex work that needs careful judgment. Sonnet 5.5, it says, is strongest on well-scoped everyday tasks: fixing bugs, and producing polished documents, slides, and spreadsheets, with what it calls a sharp eye for design. Claude Haiku 5.5, aimed at high-volume and cost-sensitive applications, is promised in the coming weeks.

Same sticker, fewer tokens

List price matches Sonnet 5: $2 per million input tokens, $10 per million output tokens, and $0.20 per million tokens for cache reads. Anthropic says Sonnet 5.5 typically needs far fewer tokens for the same work, and that in its testing the model costs up to 30% less per task than Sonnet 5. It also calls Sonnet 5.5 the fastest Sonnet to date, generating output 30%+ faster than Sonnet 5.

Against Opus 5.5, the product page lists a lower sticker. Opus 5.5 is $4 per million input tokens and $20 per million output, with cache writes at $5 per million versus $2.50 for Sonnet 5.5. Cache reads are $0.20 on both. Platform docs split the cache-write line further: $2.50 per million tokens for a 5-minute write and $4 per million for a 1-hour write. They also list a 50% Batch API discount on input and output, and a minimum cacheable prompt length of 512 tokens.

The Claude API model ID is claude-sonnet-5-5. Platform docs list a 1 million token context window, a 128K maximum output on the synchronous Messages API, adaptive thinking, a default effort of high on the Claude API, comparative latency of “Fast,” and a June 2026 reliable knowledge cutoff. Training-data cutoff is also listed as June 2026. Retirement is not sooner than September 28, 2027.

Anthropic says Sonnet 5.5 is available with zero data retention, as with Opus 5.5 and Sonnet 5, and that it is on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The platform overview names the Claude API, Amazon Bedrock (anthropic.claude-sonnet-5-5), Google Cloud, Microsoft Foundry, and Claude Platform on AWS. Those IDs are claude-sonnet-5-5 except on Bedrock.

Effort is the other cost lever. In Claude Code and Claude apps, the default is Medium. On the Claude Platform it is High. Anthropic says lower settings answer faster and use fewer tokens; higher settings reason longer and check work more thoroughly.

The scores Anthropic is publishing

Anthropic’s announcement table lists Terminal-Bench 4.0, an agentic coding evaluation, at 70.6% for Sonnet 5.5, against 10.3% for Sonnet 5 and 66.4% for Opus 5.5. Footnote 1 says the Opus 5.5 Terminal-Bench figure is at Xhigh effort, that model’s highest score. On GDPval-AA v2.1, Sonnet 5.5 is listed at 1844, two points under Opus 5.5 at 1846 and above Sonnet 5 at 1449. Footnote 3 says Artificial Analysis ran GDPval-AA and AA-Briefcase on a pre-release Claude Platform deployment that had a structured-output bug. Anthropic expects any effect to be small and to understate Sonnet 5.5, and says the bug has since been fixed.

The same table lists CursorBench 4.0 at 55.5%, against 34.1% for Sonnet 5 and 57.8% for Opus 5.5. Anthropic says that on several evaluations Sonnet 5.5 at Max effort can perform comparably to Opus 5.5, and that in its own testing and in external testing Opus 5.5 remains clearly stronger at complex, open-ended work that needs sustained judgment. Benchmark scores, it adds, capture only one facet of capability. It also says Sonnet 5.5 is the first Sonnet to beat Pokémon Red working only from screenshots.

On its cost-versus-score charts, Anthropic says Sonnet 5.5 at Low or Medium effort beats Sonnet 5’s best score on several benchmarks for about a tenth of the cost per task, and that it complements Opus 5.5 best at lower effort. Evaluation methodology is deferred to the Sonnet 5.5 system card linked from the product page.

Five breaking changes from Sonnet 5

Platform docs say five breaking changes affect code already running on Claude Sonnet 5:

Turning off up-front thinking now uses between_tools. The product page says anyone running Sonnet with thinking off must switch to that setting before moving to Sonnet 5.5. It keeps up-front thinking off and works at high effort or below.
Forced tool use returns an error.
Thinking blocks are tied to the model and the conversation.
On the Claude API and Google Cloud, the earlier computer_20251124 computer-use tool is not accepted.
The advisor tool rejects Claude Opus 4.8, Claude Opus 4.7, and Claude Sonnet 5 as advisors.

A further change does not fail requests, but it changes the response shape: text between tool calls comes back in thinking blocks. An application that streams that text to users goes quiet between tool calls until it sets a display value that returns the text, or turns off up-front thinking with between_tools. Setting temperature, top_p, or top_k to a non-default value returns a 400 error. On the Message Batches API, Sonnet 5.5 supports up to 300K output tokens with the output-300k-2026-03-24 beta header.

Safeguards, stated at the launch level

Anthropic says Sonnet 5.5 does not advance the frontier of its models’ capabilities. Because its cybersecurity capabilities are comparable to Opus 5’s, it is the first Sonnet launched with cyber safeguards and fallbacks like those on Anthropic’s most capable models. The safeguards section says those controls are similar to the ones on Opus 5.5. Higher-risk cybersecurity tasks visibly fall back to Sonnet 5. Routine software development is unaffected. Biology safeguards are the same as Sonnet 5’s and target a narrow set of high-risk requests; Anthropic says most life sciences work is unaffected.

It is also the first Sonnet to launch with safety classifiers meant to prevent reasoning extraction, and it expands preserved thinking so thinking cannot be decoupled from the account that created it. Anthropic says most developers will not notice that change.

Claude Code 2.1.284 makes it the default Sonnet

Claude Code 2.1.284, published September 28, 2026, added claude-sonnet-5-5 and made it the default Sonnet model on the Anthropic API. The release notes repeat the 1M context window and the $2 / $10 per million token price with $0.20 per million token cache reads.

Two other changes in that release matter in day-to-day use. Auto mode now offers “Yes, but ask again next time” before a read outside the working directories, so one read can be allowed without approving later ones. Claude apps gateway spend limits in /usage and the status line show dollar amounts when the gateway runs this version or later — the notes’ example is “$271.40 / $500.00 spent this month” — and rate_limits.spend_limit gains used_usd, limit_usd, and period. The same release also changes the notice shown when a Sonnet model’s safeguards flag a message, so it explains why and offers editing and retrying.

If you are moving Sonnet 5 traffic, start with the platform overview: it lists the breaking changes and links the migration guide. Then confirm Claude Code is on 2.1.284 before treating Sonnet 5.5 as the default Sonnet, and pin claude-sonnet-5-5 if a gateway or proxy still resolves the family name to the previous model.