Term
Claude Sonnet 4.5
Claude Sonnet 4.5 (September 2025) is Anthropics Sonnet-line model tuned for agents and coding — leading on SWE-bench Verified at launch, since superseded by newer Sonnet versions.
Claude Sonnet 4.5 — explained in more detail
Claude Sonnet 4.5 was released by Anthropic on 29 September 2025 and was, at launch, the most capable model in the Sonnet line. Within Anthropics range, Sonnet is the middle tier between the compact Haiku and the flagship Opus: a model built for a strong balance of capability, speed and price, which makes it the workhorse for production workloads.
This version focused on two areas: autonomous agents and software development. Anthropic reported 77.2 percent on SWE-bench Verified — a benchmark measuring real bug-fix tasks drawn from GitHub projects — placing it at the top of the models tested at release. On OSWorld, which evaluates computer use via mouse and keyboard, the model reached 61.4 percent, compared with 42.2 percent for the four-month-older Sonnet 4. Anthropic also highlighted that the model could stay on complex, multi-step tasks autonomously for around 30 hours.
Access is proprietary: Sonnet 4.5 runs through the Anthropic API, through Claude.ai and via cloud platforms such as Amazon Bedrock and Google Vertex AI. The weights are not public. On pricing it stayed at the usual Sonnet level of roughly 3 US dollars per million input and 15 US dollars per million output tokens.
Example / Practical context
In practice, Sonnet 4.5 matters mainly as the engine behind coding agents and developer tools. A typical use: an agent is given a repository and a ticket, reads through the code, edits several files, runs tests and iterates on its own until the task is done. The version was designed precisely for such running agent loops — many tool calls and long context. In Claude Code, Anthropics CLI agent, Sonnet 4.5 was for a long time the default model for day-to-day development work.
Distinction from related terms
Within the Claude family, Sonnet sits between Haiku (faster, cheaper, weaker) and Opus (stronger, pricier, slower). The 4.5 designation marks an intermediate step: an evolution of Sonnet 4 rather than a full generational jump. As of September 2026, Sonnet 4.5 is no longer the current model — Anthropic has continued the Sonnet line with newer versions (including Sonnet 5, June 2026) that supersede it in capability and benchmarks. API-only models like this differ fundamentally from open-weight models (such as Llama, Qwen or DeepSeek), whose weights can be downloaded and run locally, whereas Claude is available solely as a hosted service.