Claude Opus 4.6 with Thinking is Anthropic flagship model released in April 2026. It tops the LMArena leaderboard with an Arena Elo of 1506 and achieves 80.8% on SWE-bench Verified, making it the best model for agentic coding tasks. It features 1M token context with 76% recall on MRCR v2 (8-needle), adaptive thinking with effort controls, and 128K max output tokens. Opus 4.6 excels at complex reasoning, long document analysis, and autonomous multi-step workflows via Claude Code.
| Benchmark | Score |
|---|---|
| SWE-bench Verified | 80.8% |
| MRCR v2 (1M, 8-needle) | 76% |
| Terminal-Bench 2.0 | 65.4% |
| Arena Elo (LMArena) | 1506 |
Daily tokens (prompt + completion) routed to Claude Opus 4.6 on OpenRouter, from Jun 20 to Sep 2, from our own stored daily snapshots. Reflects OpenRouter API demand only. See all models →
| Date | Prompt /MTok | Completion /MTok |
|---|---|---|
| 2026-07-29 | $5 → $2.5 | $25 → $12.5 |
| 2026-08-01 | $2.5 → $5 | $12.5 → $25 |
| 2026-08-07 | $5 → $2.5 | $25 → $12.5 |
Changes observed across 76 daily OpenRouter catalog snapshots since Jun 19, 2026.