Anthropic releases Claude Sonnet 5.5 at Sonnet 5's price

Anthropic has released Claude Sonnet 5.5, the mid-tier model in its new Claude 5.5 family, and kept the API price exactly where Sonnet 5 left it: $2 per million input tokens and $10 per million output tokens [1]. The company’s claim is that the same work now costs less because the model uses fewer tokens and fewer tool calls, not because tokens got cheaper [1][3]. In its own testing, Anthropic says, Sonnet 5.5 costs up to 30% less per task and generates output 30%+ faster than Sonnet 5 [1]. For anyone who pays for Claude by the token, that is a claim worth checking on real work before it goes into a budget.

What was announced

Sonnet 5.5 is the second model in the Claude 5.5 family, after Opus 5.5 [1]. VentureBeat notes that Opus 5.5 was released just last week [3], and TechCrunch notes that Sonnet 5 was announced about three months ago [2]. Anthropic describes Sonnet 5.5 as “a faster, lower-cost complement” to Opus 5.5 [1]. Where Opus 5.5 is built for complex work requiring careful judgment, the company says, Sonnet 5.5 is strongest at well-scoped everyday tasks, fixing bugs, and creating documents, slides and spreadsheets [1]. A third model, Claude Haiku 5.5, built for high-volume and cost-sensitive use, will join the family “in the coming weeks” [1]. TechCrunch reports that Anthropic did not give a firm date [2].

The price is unchanged. Sonnet 5.5 costs $2 per million input tokens, $10 per million output tokens and $0.20 per million tokens for cache reads, the same as Sonnet 5 [1]. VentureBeat adds that cache writes cost $2.50 per million tokens [3]. The company says the model “typically needs far fewer tokens to do the same work” [1]. VentureBeat frames the pitch the same way: the saving comes primarily from fewer tokens and fewer tool calls rather than a lower sticker price [3]. VentureBeat also reports that it asked Anthropic how it measured the speed gain and was waiting for an answer [3].

Anthropic published a benchmark table comparing Sonnet 5.5 with Sonnet 5 and Opus 5.5 [1]. On Terminal-Bench 4.0, an agentic coding test, Sonnet 5.5 scores 70.6%, against 10.3% for Sonnet 5 [1]. On CursorBench 4.0, built from real Cursor coding sessions, it scores 55.5%, against 57.8% for Opus 5.5 [1]. On OSWorld 2.1, a computer-use test, it scores 80.1% against 81.8% for Opus 5.5 [1]. On GDPval-AA, which tests real-world tasks across 44 occupations, Anthropic says Sonnet 5.5 scores nearly level with Opus 5.5 [1]. All of these are the company’s own reported results or results it commissioned. Anthropic itself says benchmark scores capture only one facet of a model, and that in its testing Opus 5.5 “remains clearly stronger at complex, open-ended work requiring sustained judgment” [1].

The cost-per-task claim rests on charts that plot score against cost at each effort level [1]. Anthropic says that on several benchmarks, Sonnet 5.5 at Low or Medium effort beats Sonnet 5’s best score for about a tenth of the cost per task [1]. The effort setting matters for what you pay: in Claude Code and the Claude apps the default is Medium, while the Claude Platform defaults to High [1]. Lower settings answer faster and use fewer tokens; higher settings reason longer and check work more thoroughly, the company says [1].

Customer quotes on the launch page point the same way, and they are vendor-selected. Slack’s Curtis Allen says Sonnet 5.5 did better than Sonnet 5 on almost all of its offline Slackbot evals, with about 14% fewer output tokens [1]. Zendesk’s Abhinay Kathuria says tickets were processed 20% faster [1]. Lovable’s Fabian Hedin says its coding evals showed a third fewer tool calls and roughly half the shell runs [1]. Anthropic also lists a finance customer whose private suite of 2,441 tasks saw Sonnet 5.5 use about 121k tokens per answer, against 497k for Sonnet 5 [1]. None of these tests can be reproduced from the public page.

What changed, and for whom

The biggest behavioural change is on security work. Because Sonnet 5.5’s cybersecurity capabilities are comparable to Opus 5’s, Anthropic says it is the first Sonnet model to launch with cyber safeguards and fallbacks like those on its most capable models [1]. TechCrunch reports the same: Sonnet 5.5 is the first Sonnet subject to the cyber safeguards that apply to Fable and Opus [2]. In practice, Anthropic says, users can still find and fix bugs in their code as part of routine development, but “higher-risk cybersecurity tasks will visibly fall back to Sonnet 5” [1]. Cyber defenders will soon be able to apply to an expanded Cyber Verification Program for tiered access to more advanced capabilities [1]. Biology safeguards stay the same as Sonnet 5’s; the company says some microbiology and virology requests may be flagged in error [1].

The second change is for developers. If you run Sonnet with thinking off, Anthropic says you need to switch to a new between_tools setting, which keeps up-front thinking off, before moving to Sonnet 5.5 [1]. The model also ships with classifiers meant to stop reasoning extraction, and it expands preserved thinking so Claude’s thinking cannot be decoupled from the account that created it [1]. The company says most developers will not notice; if you move conversations between accounts, including switching accounts mid-session in Claude Code, its docs explain the change [1].

Availability is broad. Sonnet 5.5 is available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, and it is offered with zero data retention, as Opus 5.5 and Sonnet 5 are [1].

The competitive picture explains the pricing choice. VentureBeat reports that OpenAI’s GPT-6 Sol lists Standard API pricing at the same $2 per million input tokens and $10 per million output tokens [3]. It also reports that Google offers Gemini 3.8 Flash at an introductory $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, with standard pricing rising to $1.50 and $7.50 on January 1, 2027 [3]. Anthropic is not trying to win on the cheapest tokens; it is arguing that a model that finishes in fewer steps costs less to run [3].

Who it is for — and not

Sonnet 5.5 is aimed at people and teams who run a lot of well-defined work through Claude: bug fixes, code changes in a known codebase, reports, decks and spreadsheets. If most of your Claude use looks like that, the promise is the same result for fewer tokens and less waiting [1]. Teams building agents that call many tools may see the largest effect, because the customer tests Anthropic quotes focus on fewer steps and fewer tool calls [1].

It is not a replacement for Opus 5.5 on hard, open-ended problems, and Anthropic says so directly [1]. If your work depends on long chains of judgment, such as a strategy memo with ambiguous inputs or a design decision with no clear right answer, the company’s own guidance is to keep Opus 5.5 for it [1].

It also needs care from anyone doing security work. Penetration testers, red teams and researchers who used Sonnet 5 for higher-risk tasks should expect some requests to fall back to Sonnet 5 and plan for that, or apply to the Cyber Verification Program when it expands [1].

Finally, it is not a price cut. If your bill is dominated by a workload where Sonnet 5.5 happens not to save tokens, your costs stay the same [1][3]. The only way to know is to measure your own tasks.

This page will be updated if Anthropic changes Sonnet 5.5’s pricing or safeguards, when Haiku 5.5 ships with a public date, and if independent tests confirm or contradict the cost-per-task claim.

∴ Same sticker price, fewer tokens per task: re-test your Sonnet workloads before assuming the bill drops.