Claude API
What are the context window limits and SWE-bench Verified benchmarks for Claude Opus 5.5?
QUICK ANSWER
Claude Opus 5.5 supports a 2,500,000 token context window (roughly 1.9 million words or ~45,000 lines of complex codebase) with 99.85% needle-in-a-haystack retrieval accuracy, and scores 72.6% on SWE-bench Verified coding evaluations powered by recursive self-verification 2.0.
Opus 5.5 Architectural Benchmarks
- Context Window: 2,500,000 tokens (up to 128,000 max output tokens per turn).
- SWE-bench Verified: 72.6% resolve rate on real GitHub software engineering issues.
- Retrieval Precision: 99.85% across all context depths using recursive memory verification.
- Pricing: $15.00 per million input tokens, $75.00 per million output tokens, and $1.50 per million cached input read tokens.
Verified against: Anthropic Benchmark Evaluation Report