Introducing Claude Opus 4.7
Our most capable model ever, with unprecedented reasoning, creative depth, and a 1M native context window.
April 1, 2026 · 6 min read
Today we're excited to announce Claude Opus 4.7, the newest addition to the Claude model family. Opus 4.7 represents a fundamental leap in AI capability, achieving new state-of-the-art results across every major reasoning, coding, and creative benchmark we track.
Opus 4.7 is available immediately on claude.ai for all Pro and Team subscribers, and via the API as claude-opus-4-7-20260401.
Benchmark Results
Opus 4.7 sets new records across the board:
| Benchmark | Opus 4.6 | Opus 4.7 |
|---|---|---|
| GPQA Diamond | 74.9% | 89.2% |
| MATH-500 | 96.4% | 99.1% |
| SWE-bench Verified | 72.5% | 84.7% |
| HumanEval+ | 93.0% | 97.8% |
| MMLU-Pro | 84.1% | 91.6% |
| TAU-bench (Airline) | 53.0% | 71.4% |
What's New
Beyond raw benchmark performance, Opus 4.7 introduces several notable improvements:
1M Native Context Window. Opus 4.7 ships with a 1,000,000 token context window standard across all tiers — no special access required. Internal evaluations show near-perfect recall across the full window.
Persistent Memory. Opus 4.7 can now maintain a structured memory graph across conversations, allowing it to remember user preferences, past decisions, and project context without re-prompting.
Native Image Generation. For the first time in a Claude model, Opus 4.7 can generate and edit images natively within the conversation, powered by a new multimodal architecture.
We believe Opus 4.7 represents an important step toward our mission of building AI systems that are safe, beneficial, and understandable. We can't wait to see what you build with it.