Anthropic releases Claude Sonnet 4.6
Early testers preferred it to Sonnet 4.5 on coding tasks about 70% of the time, and to the larger Opus 4.5 about 59% of the time, at unchanged Sonnet pricing.
- Models & capabilities
- Notable
Anthropic released Claude Sonnet 4.6, which it described as a full upgrade of the Sonnet line’s skills across coding, computer use, long-context reasoning, agent planning, knowledge work and design, rather than a narrower point update. The model shipped with a 1 million-token context window in beta, large enough, Anthropic said, to hold an entire codebase, a lengthy contract, or several research papers at once.
Anthropic reported that early testers preferred Sonnet 4.6’s coding output to Sonnet 4.5’s roughly 70% of the time, and — more notably — preferred it to the company’s larger and more expensive Opus 4.5 model 59% of the time on the same tasks. On the OSWorld computer-use benchmark the model showed a substantial jump over earlier Sonnet versions, part of a broader push by Anthropic to close the gap between its mid-tier and flagship models. Pricing was left unchanged from Sonnet 4.5, at $3 per million input tokens and $15 per million output tokens on the API.
The company also reported gains in consistency and instruction-following, with fewer false claims of task completion and better handling of multi-step agentic work — the categories of failure that had most limited Sonnet’s use in unsupervised agent deployments.
The release continued Anthropic’s practice of shipping Sonnet as the workhorse tier below Opus: Opus 4.6 had arrived less than two weeks earlier, and later Opus updates — 4.7 in April and 4.8 in May — continued the pattern of rapid, closely spaced point releases across both model lines.