Claude Opus 4.7
Claude Opus 4.7 raises SWE-bench Pro to 64.3% and accepts images over 3x larger in pixels (2,576 px), while deliberately limiting cyber capability.
Stronger software engineering and instruction following, high-resolution vision (about 3.75 megapixels) and file-system memory. New tokenizer maps text to roughly 1.0-1.35x more tokens. Cyber capabilities intentionally below Mythos Preview, with automatic blocking and a Cyber Verification Program for security professionals. Price unchanged at $5/$25.
- Date
- Thursday, 16 April 2026
- Lab
- Anthropic
- Kind
- model
- Access
- closed API
- Price
- $5/$25 per M tokens in/out, unchanged from Opus 4.6
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-bench Pro | 64.3% vs 53.4% for Opus 4.6, 57.7% GPT-5.4, 54.2% Gemini 3.1 Pro; vendor-reported table per Vellum | company |
| SWE-bench Verified | 87.6% vs 80.8% for Opus 4.6; Mythos Preview 93.9% | company |
| XBOW visual-acuity benchmark | 98.5% vs 54.5% for Opus 4.6 | third-party |
| CursorBench | 70% vs 58% for Opus 4.6, measured by Cursor | third-party |
Anthropic's own benchmark table is an image; the SWE-bench Pro, SWE-bench Verified and comparator numbers come from a third-party transcription of that vendor table (Vellum) and are flagged vendor-reported, while the CursorBench, XBOW and image-size figures are in Anthropic's text. Anthropic says the model accepts images up to 2,576 px on the long edge (about 3.75 MP), more than three times the pixels of prior Claude models. OpenAI shipped GPT-5.5 a week later (2026-04-23, TechCrunch). The API-wide 'xhigh' effort level and task budgets arrived with this model.
Sources
- www.anthropic.com/news/claude-opus-4-7
- www.vellum.ai/blog/claude-opus-4-7-benchmarks-explained
- techcrunch.com/2026/04/23/openai-chatgpt-gpt-5-5-ai-model-superapp/
This record was checked against its sources on 6 October 2026. How we check