Anthropic: Claude Sonnet 4

For developers, the persistent annoyance is watching an assistant lose the thread of a large codebase or misread an ambiguous instruction. Sonnet 4 is built to remove that friction, improving on Sonnet 3.7 with more precise, controllable coding and reasoning and posting a state-of-the-art 72.7% on SWE-bench. A representative task is pointing it at a sprawling repository, loading the relevant files into its 1,000,000-token context window, and having it reason through a multi-file change while calling tools to run and verify the edits. Vision input lets it read screenshots or design mockups in the same session, and its knowledge cutoff of 2025-01-31 keeps it reasonably current. Against the broad Claude, ChatGPT and Gemini assistants it is compared with, this is a specific engineering-focused release rather than a consumer chat app, and unlike the pricier Opus tier it aims at a lower cost point, listed at $3 per million input tokens, still above average but below the flagship. Ideal for research and writing teams and engineers who need controllability and a large working memory without paying top-tier rates. It fits users who value benchmark-backed coding accuracy over the absolute cheapest option. Choose it for demanding coding and reasoning work on a mid-premium budget; look elsewhere if price is your only constraint.