Anthropic launches Claude Opus 4.7 with higher-res vision

Anthropic introduced Claude Opus 4.7 on April 16, 2026. The company describes the release as a general‑availability upgrade focused on software engineering, long‑horizon workflows and financial modeling. Opus 4.7 replaces the February 4.6 build and is available through the Claude API and on Amazon Bedrock, Google Vertex AI and Microsoft Foundry.

Anthropic reported the model scored 87.6% on the SWE‑bench Verified coding benchmark, up from 80.8% on Opus 4.6. On the company’s General Finance module, Opus 4.7 posted a 0.813 result, compared with 0.767 for 4.6. Internal evaluations also showed the model outperforming competing models on tests of tool use and computer interaction.

A major technical change is vision: Opus 4.7 accepts images with a long edge up to 2,576 pixels, three times the previous limit. Anthropic said the larger images help the model read complex charts, interfaces and technical diagrams more accurately, and warned that higher‑resolution inputs use more tokens and can increase cost for image‑heavy workloads.

The release adds developer features inside the Claude Code environment. A /ultrareview tool lets professional and max‑tier users run multi‑agent sessions to find bugs and design problems. A task budget option entered public beta to help control spending during long autonomous runs. The model maintains the 1 million token context window introduced earlier in the year.

Developers testing Opus 4.7 reported that the model follows instructions more literally than earlier versions, which can require reworking prompts that were tuned for older releases. Anthropic acknowledged an updated tokenizer can consume between 1.0 and 1.35 times more tokens for the same input compared with 4.6. Pricing for Opus 4.7 remains $5 per million input tokens and $25 per million output tokens.

Anthropic said it built new automated safeguards into Opus 4.7 to block high‑risk cybersecurity uses and to reduce harmful outputs. The company also noted improved honesty metrics in its system documentation. Anthropic continues to test a separate, restricted Mythos Preview for cybersecurity applications.

Early community reaction included a post on X describing the update as higher capability requiring more careful prompt work: “Claude 4.7 is out, and using it feels like stepping into an F1 car. Far more power, and it does exactly what you tell it at full speed. Your job is to pick the direction and make the turns,” a user wrote.

Anthropic framed the release for commercial use and emphasized the model is intended for autonomous workflows and complex technical tasks. Opus 4.7 is available now to customers through Anthropic’s API and partner cloud platforms.