Claude Sonnet 4.6 vs Sonnet 4.5: Full Comparison
Sonnet 4.6 is the latest upgrade to the Sonnet line, replacing 4.5 as the default model for Claude Code and claude.ai. The biggest upgrade is adaptive thinking, previously exclusive to Opus. Sonnet 4.6 achieves a 70% user preference win rate over 4.5 and scores 80.2% on SWE-bench. Same pricing at $3/$15 per MTok.
| Spec | Claude Sonnet 4.6 | Claude Sonnet 4.5 |
|---|---|---|
| Context Window | 200K | 200K |
| Max Output | 64K | 64K |
| Vision | ||
| Extended Thinking | ||
| Adaptive Thinking | ||
| Input Price | $3/MTok | $3/MTok |
| Output Price | $15/MTok | $15/MTok |
Sonnet 4.6 is a direct upgrade over 4.5 at the same price. With adaptive thinking, higher benchmarks, and 70% user preference, there is no reason to stay on 4.5 for new projects. Migrate to Sonnet 4.6 for better results.
What is new in Claude Sonnet 4.6 compared to 4.5?
Sonnet 4.6 adds adaptive thinking (previously Opus-only), achieves SWE-bench 80.2% (up from ~70%), ARC-AGI-2 60.4%, and scores 72.5% on OSWorld. Users prefer it 70% of the time over 4.5. Pricing remains the same at $3/$15 per MTok.
Should I migrate from Sonnet 4.5 to Sonnet 4.6?
Yes. Sonnet 4.6 is a direct upgrade at the same price point. Simply change your model ID from claude-sonnet-4-5-20250929 to claude-sonnet-4-6. The API is fully compatible — no code changes needed beyond the model ID.