Claude Opus 4.7
Anthropic·
LLMsflagshipcloud
Overview
Anthropic's most capable generally available model. Default in Claude Code. Dense architecture with new tokenizer, file-system-based cross-session memory, new 'xhigh' effort level, and task budgets (public beta) for agentic loops. 1M context window with no long-context premium. Sits below Mythos Preview in capability, above Opus 4.6. $5/M input, $25/M output tokens.
Capabilities and innovations
1M context windowVision input up to 3.75 MPComputer use with 1:1 pixel mappingAgentic workflowsCross-session file-system memoryxhigh Effort LevelTask Budgets (Public Beta)File-System MemoryNew Tokenizer
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 94.2% | unsourced |
| SWE-bench Verified | 87.6% | unsourced |
| SWE-bench Pro | 64.3% | unsourced |
| MMLU-Pro | 89.87% | unsourced |
| Terminal-Bench 2.x | 68.54% | independent Terminal-Bench 2.1 68.54%, measured by Vals AI with the Terminus 2 harness (accessed 2026-09-23). Vals AI is the reference harness for this field so values compare across vendors; vendor and other third-party figures stay in the notes below. Replaces 69.4 (unsourced). |
- Humanity’s Last Exam (no tools): not reported by the vendor
- MMLU: not reported by the vendor
- MMMU-Pro: not reported by the vendor
Reliability
- Hallucination rate (Vectara HHEM)
- 12% (lower is better)independent
Links
More from Anthropic
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.