Claude Sonnet 4

Claude Sonnet 4 (API name claude-sonnet-4-20250514) launched on 22 May 2025 alongside Claude Opus 4 and did something the flagship did not: it scored slightly higher than its bigger sibling on SWE-bench Verified — 72.7% against 72.5% — while costing a fifth as much, at $3 per million input tokens and $15 per million output. The gap between the two models was real in long, open-ended agent work, where Opus 4 sustained focus for hours; on a single well-specified coding task the cheap model was already at the frontier. That inversion is why Sonnet became the default choice for most production traffic in this generation rather than the compromise option. It was also the only model of the pair offered to free Claude users, which put frontier-class coding in front of people who were not paying for it. GitHub adopted it as the model behind the then-new Copilot coding agent, and iGent reported navigation errors in autonomous multi-file development dropping from roughly 20% to near zero. In August 2025 Anthropic extended Sonnet 4 to a one-million-token context window in public beta — five times the 200,000 tokens it shipped with, and enough to hold an entire mid-sized codebase in one request. Long-context requests were billed at a premium rate. For most of its life the model ran with 200,000 tokens of context and up to 64,000 output tokens. Anthropic notified developers on 14 April 2026 and retired both Claude 4 models on 15 June 2026. The recommended replacement is claude-sonnet-4-6.

▸Videos

No videos yet.

▸Reviews