Anthropic Releases Claude Sonnet 4.6: Near-Opus Performance at $3/MTok

Closing the Gap to Opus

Anthropic released Claude Sonnet 4.6 on February 17, 2026, positioning it as a full-capability upgrade at the Sonnet price tier. The model achieves 72.5% on OSWorld-Verified—a benchmark measuring autonomous computer operation—compared to 72.7% for Opus 4.6. That 0.2-point gap is effectively negligible, while Sonnet 4.6 outperforms GPT-5.2 by 34.3 points on the same benchmark.

Computer Use Progress

Sonnet 4.6 represents a steep 16-month arc of improvement in computer use: from 14.9% (Sonnet 3.5) to 28.0% (Sonnet 3.5 v2), 42.2% (Sonnet 3.6), 61.4% (Sonnet 4.5), and 72.5% today. In real-world testing on insurance-specific workflows—form filling, data entry, multi-step web applications—Anthropic reports 94% task accuracy. The model can navigate spreadsheets, complete web forms, and operate GUI applications with near-human fluency.

1M Token Context Window (Beta)

Sonnet 4.6 ships with a 1 million token context window in beta, doubling the previous Sonnet maximum. This is sufficient to load entire large codebases, lengthy legal contracts, or hundreds of research papers into a single request without chunking or retrieval.

Pricing and Availability

Pricing holds at $3 input / $15 output per million tokens—identical to Sonnet 4.5. It is now the default model on Free and Pro tiers of claude.ai and Claude Cowork, and is available across the Claude API, AWS Bedrock, and Google Cloud Vertex AI.

Source: Anthropic