Anthropic released Claude Sonnet 4.6 on February 17, 2026. The headline: it performs within 1-2 points of Opus 4.6 on key benchmarks — at a fraction of the price. Early testers preferred it over Sonnet 4.5 about 70% of the time.
Key changes
| Sonnet 4.5 | Sonnet 4.6 | |
|---|---|---|
| Release date | Late 2025 | February 17, 2026 |
| Context window | 200K tokens | 1M tokens |
| SWE-bench Verified | 77.2% | 79.6% |
| OSWorld (computer use) | 61.4% | 72.5% |
| Adaptive thinking | No | Yes |
| Preferred over 4.5 | — | 70% of the time |
| Preferred over Opus 4.5 | — | 59% of the time |
| Input price | $3 / 1M tokens | $3 / 1M tokens |
| Output price | $15 / 1M tokens | $15 / 1M tokens |
What improved
Context window: 200K → 1M tokens. Same jump as Opus 4.6 — you can now process massive codebases and documents at Sonnet pricing.
Coding (SWE-bench): 77.2% → 79.6%. A meaningful improvement. Testers noted better context reading before modifying code and reduced logic duplication.
Computer use (OSWorld): 61.4% → 72.5%. A huge 18% jump in autonomous computer use — relevant if you’re building agents that interact with UIs.
Adaptive thinking. Like Opus 4.6, Sonnet 4.6 dynamically adjusts reasoning depth based on task complexity.
It beats Opus 4.5. Users preferred Sonnet 4.6 over the previous flagship Opus 4.5 in 59% of head-to-head comparisons. That’s remarkable for a model at one-fifth the price.
What stayed the same
- Pricing ($3/$15 per million tokens)
- API compatibility
Should you upgrade?
Yes. Same price, better at everything, bigger context window. Sonnet 4.6 is now the default model on claude.ai — Anthropic clearly considers it the go-to choice for most users.
See our full AI Model Comparison for all models side by side.