Anthropic Releases Claude Opus 4.7 With Major Coding and Vision Upgrades

Anthropic today released Claude Opus 4.7, its most capable generally available model to date. Available immediately across the Anthropic API, all Claude products, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, Opus 4.7 brings meaningful gains in software engineering, vision, and instruction following -- while maintaining the same pricing as its predecessor at $5 per million input tokens and $25 per million output tokens.

The most significant leap is in coding. Anthropic reports that Opus 4.7 lifts resolution by 13% on its internal 93-task software engineering benchmark compared to Opus 4.6, including solving four tasks that neither Opus 4.6 nor Sonnet 4.6 could complete. In practice, the model writes cleaner code -- avoiding meaningless wrapper functions and fallback scaffolding -- and actively verifies its own output before reporting results. For vision, Opus 4.7 accepts images up to 2,576 pixels on the long edge, more than three times the resolution of prior Claude models, making it substantially more capable at tasks like computer-use agents, technical diagram analysis, and data extraction from dense visuals.

Instruction following has also been overhauled. Opus 4.7 interprets prompts more literally and precisely, which Anthropic warns may produce different behavior for prompts written for earlier models. The model also introduces a new xhigh effort setting, giving developers finer control over the tradeoff between reasoning depth and latency. On efficiency, Anthropic notes that Opus 4.7 running at low effort is roughly equivalent to Opus 4.6 at medium effort -- a meaningful improvement for cost-sensitive agentic workloads.

One notable aspect of the release is what Anthropic chose to say publicly about Mythos -- the company's most advanced model, which remains unreleased due to safety concerns. Anthropic acknowledged that Opus 4.7, despite being its best publicly available model, is "less broadly capable" than Mythos Preview. The comparison is unusual candor from an AI lab and signals that Anthropic is actively managing a gap between its frontier research and what it considers safe to deploy. For developers, Opus 4.7 is available today via the claude-opus-4-7 model identifier across all major cloud platforms.