Claude Opus 4.7 arrives with better vision, memory, and instruction-following

Anthropic has released Claude Opus 4.7, a direct upgrade to Opus 4.6, now available across Claude products, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at the same price point. Key improvements include more literal instruction-following (which may break prompts written for older models), better vision supporting images with 3x more pixels, enhanced file system-based memory, improved creativity for professional tasks, and stronger finance analysis performance. Safety metrics are largely unchanged from Opus 4.6, with one noted regression in harm-reduction advice for controlled substances. Anthropic also warns that Opus 4.7 uses an updated tokenizer and thinks more at higher effort levels, which may increase token consumption. The model is positioned below the more capable Claude Mythos Preview and serves as a testing ground for new cybersecurity safeguards before broader Mythos-class model releases.