Claude Opus 4.7, released on April 16, 2026, is one of the Anthropic's publicly available AI model in the Claude 4 family. It represents a significant leap in coding, agentic workflows, and visual understanding, with benchmark-leading performance on software engineering and multi-step tool-calling tasks.
- Elite Coding Performance: Achieves 87.6% on SWE-bench Verified and 64.3% on SWE-bench Pro, representing a nearly 7-point and 10.9-point improvement over Opus 4.6 respectively.
- 1M Token Context Window: Full million-token context at standard pricing with no long-context premium, enabling analysis of entire codebases and lengthy documents.
- High-Resolution Vision: Accepts images up to 2,576 pixels on the long edge (~3.75 megapixels), over 3x the resolution of prior Claude models, with improved understanding of chemical structures and technical diagrams.
- Task Budgets: A new feature giving the model a rough estimate of how many tokens to target for a full agentic loop, including thinking, tool calls, tool results, and final output.
- xhigh Effort Level: A new effort level between high and max, providing finer control over the quality-speed-cost tradeoff. This is the default for Claude Code subscribers.
- Production Cybersecurity Safeguards: Ships with built-in security protections tested on Opus before rolling out to Mythos-class models.
- Autonomous Software Engineering: Handles complex, long-running coding tasks with rigor and consistency, devising ways to verify its own outputs before reporting back. Ideal for large-scale refactors, bug resolution, and feature implementation.
- Multi-Step Agentic Workflows: Excels in orchestrating sequences of tool calls, API interactions, and decision-making loops with precise instruction-following — suitable for production agent deployments.
- Visual Document Analysis: The 3.3x resolution upgrade makes it highly effective for reading technical diagrams, charts, screenshots, chemical structures, and complex visual documents.
- Long-Context Research: With 1M tokens of context and 128K max output, well-suited for synthesizing information across entire repositories, legal corpora, or research paper collections.
| Capability | Description |
|---|
| Reasoning | GPQA Diamond: 94.2%. Strong analytical and scientific reasoning with adaptive thinking support. |
| Coding | SWE-bench Verified: 87.6%, SWE-bench Pro: 64.3%, Terminal-Bench: 69.4%. |
| Agentic | MCP-Atlas scaled tool use: 77.3%, Finance Agent v1.1: 64.4%. Rigorous multi-step execution. |
| Computer Use | OSWorld-Verified: 78.0%. Strong desktop and browser interaction capabilities. |
| Multimodal | Text and image input. High-resolution vision up to ~3.75 MP. CharXiv: 82.1% (91.0% with tools). |
| Context Window | 1,000,000 tokens input. |
| Max Output | 128,000 tokens standard; 300,000 tokens via Message Batches API with beta header. |
| Tool Use | Full function calling, code execution, MCP support, adaptive thinking, task budgets. |
| Multilingual | Strong multilingual performance across major world languages. |
- Claude Mythos Preview remains more broadly capable, leading Opus 4.7 on most benchmarks including SWE-bench Pro (77.8% vs 64.3%) and SWE-bench Verified (93.9% vs 87.6%).
- New tokenizer produces up to 35% more tokens for the same input text, meaning actual per-request costs may increase despite unchanged per-token pricing.
- Image input only (no audio or video input natively).
- No real-time or internet-connected capabilities without tool integrations.
| Model | Input (Credits/Token) | Output (Credits/Token) | Notes |
|---|
| Claude Opus 4.7 | 5.00 | 25.00 | Up to 90% savings with prompt caching |
