Anthropic releases Claude Sonnet 5.5, saying it is 30% faster and up to 30% cheaper per task

Anthropic announced Claude Sonnet 5.5 on 28 September 2026, calling it the second model in its Claude 5.5 family and “a clear upgrade over Claude Sonnet 5” that “runs 30%+ faster” and “costs up to 30% less for most work”.^[1][4] Business Today reported the launch on 29 September.^[5] Anthropic’s page does not give a time of day for the announcement.^[1]

Anthropic said the list price is unchanged from Sonnet 5, at $2 per million input tokens, $10 per million output tokens and $0.20 per million tokens for cache reads, and that “in our testing, it costs up to 30% less per task” because it “typically needs far fewer tokens to do the same work”.^[1][2] Claude Opus 5.5 is priced at $4 and $20.^[1][4] VentureBeat reported that it had asked Anthropic how the speed gain was measured and would update when it heard back.^[4]

On Anthropic’s own tests, Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, a coding evaluation, against 10.3% for Sonnet 5 and 66.4% for Opus 5.5, and 55.5% on CursorBench 4.0 against 57.8% for Opus 5.5.^[1][4] Anthropic said Opus 5.5 “remains clearly stronger at complex, open-ended work requiring sustained judgment”.^[1] Box, Zendesk and Slack, quoted by Anthropic from early testing, reported respectively that Sonnet 5.5 was “2.4x faster” and “used 12% fewer total tokens”, that tickets were “processed 20% faster”, and that it used about 14% fewer output tokens.^[1][4]

Artificial Analysis, an independent benchmarking firm, said on 28 September that Sonnet 5.5 at maximum effort scored 56 on its Intelligence Index, “just 2 points behind Opus 5.5 (max)”, but used “the highest Output Tokens per Task we’ve seen” and cost $7.60 per task, about 50% more than Sonnet 5.^[6] It measured 64% on Terminal-Bench 4.0.^[6] It ran its tests on a pre-release deployment that Anthropic later found had a bug affecting structured outputs, said it would re-run them, and Anthropic expects any effect on the scores to be small.^[1][6] Both Artificial Analysis and VentureBeat noted that the $2 and $10 price matches OpenAI’s GPT-6 Sol.^[4][6]

Anthropic said Sonnet 5.5 is the first Sonnet model to launch with cybersecurity safeguards like those on its most capable models, so that “higher-risk cybersecurity tasks will visibly fall back to Sonnet 5”, and with classifiers meant to block extraction of its reasoning.^[1] Its developer documentation lists five breaking changes for code already running on Sonnet 5, including that that turning thinking off with “disabled” and forced tool use now return errors.^[2] The model is available as claude-sonnet-5-5 on the Claude API, AWS, Google Cloud and Microsoft Foundry, and Anthropic said Claude Haiku 5.5 will follow “in the coming weeks”; TechCrunch noted that no firm date was given.^[1][2][3]