Anthropic has released Claude Sonnet 5.5, a new version of its mid-tier artificial-intelligence model that the company says is faster than its predecessor and particularly strong at coding. The model costs $2 per million input tokens and $10 per million output tokens, the same list price as Sonnet 5 and half the price of Anthropic’s Opus 5.5.
The launch puts Sonnet 5.5 into direct competition with similarly priced models from OpenAI, while raising questions about how much its performance costs when users allow it to spend more tokens and effort on each task.
Coding performance leads the launch
Anthropic said Sonnet 5.5 is more than 30% faster than Sonnet 5, which was released in June. The company described the new model as best suited to clearly defined everyday work, bug fixes and the creation of polished documents, presentations and spreadsheets. It also highlighted the model’s ability to assess design.
Coding is the area where Sonnet 5.5 made its strongest showing. On Terminal-Bench 4.0, a test that measures whether an AI agent can complete complex professional tasks by entering commands independently, Sonnet 5.5 achieved a 70.6% completion rate, according to Anthropic. Opus 5.5 scored 66.4%, while the earlier Sonnet 5 reached 10.3%.
The result was also positive in testing by Artificial Analysis, an independent model-testing firm. Its results gave Sonnet 5.5 a 63.6% score, ahead of Opus 5.5 at 59.6% and OpenAI’s GPT-6 Astra at 59.1%.
Artificial Analysis ranked Sonnet 5.5 second overall behind Opus 5.5, but noted that the newer Sonnet used more tokens per task than any other model it tested. Tokens are the small pieces of text that AI systems process, and providers generally charge separately for the tokens a model reads and generates.
Anthropic also reported that Sonnet 5.5 at its High effort setting matched GPT-6 Sol on the FrontierCode test at approximately one-fifth of the cost per task. Effort settings allow a model to spend more time processing a problem, potentially improving its answer while increasing usage and expense.
Price advantage comes with a usage caveat
The headline price for Sonnet 5.5 remains $2 per million input tokens and $10 per million output tokens. That is half the price of Opus 5.5. Anthropic also says Sonnet 5.5 uses nearly one-third fewer tokens per task than Sonnet 5, which can make it less expensive to operate despite the unchanged per-token rates.
However, the model’s token consumption varies substantially with its effort setting. Artificial Analysis found that, at maximum effort, Sonnet 5.5 generated about 193,000 tokens per test task. That was roughly 60% more than Opus 5.5 and the highest level measured by the testing firm. The resulting cost was $7.60 per task, approximately 50% more than Sonnet 5.
That finding complicates Anthropic’s claim of savings of up to 30%. Anthropic’s comparison uses lower settings: in its applications, Medium effort is the default, and the company says Sonnet 5.5 can exceed Sonnet 5’s best coding result for less than one-tenth of the cost. Artificial Analysis said High effort offered the best value in its testing.
On GDPval-AA, a benchmark covering professional work across 44 occupations and using an Elo-style ranking system, Sonnet 5.5 scored 1,844. Opus 5.5 scored 1,846, making the two models effectively tied. GPT-6 Sol recorded 1,487.
According to Decrypt, Anthropic’s benchmark figures are self-reported, while Artificial Analysis tested a pre-release version that contained a bug. Anthropic said the bug was expected to have little effect or to have slightly understated the model’s scores. The company maintained that Opus 5.5 remains clearly stronger for complex work requiring sustained judgment.
Competitive pricing and Moneyiar market picture
The launch comes as AI companies compete on both capability and operating cost. OpenAI reduced GPT-6 Sol’s price to $2 per million input tokens and $10 per million output tokens the previous week. OpenAI’s GPT-5.6 Terra, another mid-tier model, is listed at $2 per million input tokens and $12 per million output tokens. Anthropic did not publish Terra benchmark results.
Anthropic said Claude Haiku 5.5, intended for high-volume and cost-sensitive applications, is due in the coming weeks. No further pricing or performance details for that model were provided.
Moneyiar market data, as of the times shown below, recorded the following moves:
- Ethereum Name Service (ENS): $7.0684912163, up 0.09% over 24 hours and 12.69% over seven days, as of September 25, 2026, at 22:04 +03:30.
- Audiera (BEAT): $0.0966461527, up 28.94% over 24 hours but down 22.33% over seven days, as of September 12, 2026, at 03:59 +03:30.
- Decimal (DEL): $0.4597318509, down 2.56% over 24 hours and 4.56% over seven days, as of September 25, 2026, at 22:03 +03:30.
- Mina (MINA): $0.138728546, down 5.53% over 24 hours but up 30.48% over seven days, as of September 25, 2026, at 22:03 +03:30.
The pricing and benchmark results make Sonnet 5.5 a significant new option for coding and other structured tasks, but its actual cost depends on how much effort and token usage a user permits. Anthropic’s next model in the lineup, Haiku 5.5, is expected in the coming weeks.
