Sonnet 5.5 Costs Half, Codes Harder—Anthropic's Mid-Tier Eats Opus's Lunch 🧠

Anthropic released Claude Sonnet 5.5 on Monday, positioning the mid-tier model as a faster, cheaper alternative to its flagship Opus 5.5. Pricing holds at $2 per million input tokens and $10 per million output tokens, half of Opus 5.5's rates, while running more than 30% faster than its predecessor, Sonnet 5, which launched in June. Anthropic framed the release in direct terms: "Sonnet 5.5 is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. It's also got a sharp eye for design."

The benchmark gap is where Anthropic lands its sharpest punch. On Terminal-Bench 4.0, which scores AI agents on the share of complex professional tasks completed via typed commands, Sonnet 5.5 reached 70.6% on Anthropic's reported numbers, ahead of Opus 5.5 at 66.4% and far above Sonnet 5 at 10.3%. Independent firm Artificial Analysis corroborated the lead on its own run, recording 63.6% for Sonnet 5.5, 59.6% for Opus 5.5, and 59.1% for OpenAI's GPT-6 Astra. In a September 28 post, Artificial Analysis noted that Sonnet 5.5 (max) scored 64% on Terminal-Bench 4.0, a 50-point increase over Sonnet 5 (max) and slightly above Opus 5.5 and GPT-6.

Cost-per-task depends heavily on the effort setting, the dial that controls how long a model deliberates. Anthropic said Sonnet 5.5 at High effort matches GPT-6 Sol on FrontierCode for roughly a fifth of the cost per task, while on GDPval-AA, an Elo-based benchmark grading real-world work across 44 occupations, Sonnet 5.5 scored 1844 to Opus 5.5's 1846, an effective tie, with GPT-6 Sol trailing at 1487. The pricing pressure is industry-wide: OpenAI cut GPT-6 Sol to $2 and $10 last week, and GPT-5.6 Terra lists at $2 and $12.

Token consumption tempers the savings story. At max effort, Sonnet 5.5 wrote about 193,000 tokens per test task, the highest Artificial Analysis has measured and roughly 60% more than Opus 5.5, producing a per-task cost of $7.60, about 50% above Sonnet 5. Anthropic's projected savings apply at lower settings; at Medium effort, the default in Anthropic's apps, the company said Sonnet 5.5 beats Sonnet 5's best coding score for less than a tenth of the cost, while Artificial Analysis identified High effort as the best overall value. Anthropic's figures are self-reported, and Artificial Analysis tested a pre-release build with a known bug that the company said likely changed scores little or slightly understated them.