Claude Sonnet 4.6 hits benchmark parity with Opus

Sonnet 4.6 effectively kills the premium for frontier intelligence, offering Opus-level reasoning at mid-tier pricing for developers.

• –1M context window and "context compaction" allow for massive codebase ingestion without typical performance degradation.
• –Significant gains in computer use (72.5% OSWorld) position the model as a reliable autonomous operator rather than just a text generator.
• –Benchmark data shows it surpassing Opus 4.5 in coding and reasoning while maintaining $3/$15 per million token pricing.
• –Matt Maher's planning benchmark reveals that IDE integration (like Cursor) is now a primary driver of model performance over raw CLI usage.
• –Adaptive thinking effort controls provide a new lever for optimizing token spend against task complexity.