Claude Opus 4.7 takes a leap in autonomous coding, vision and professional work

Anthropic has published Claude Opus 4.7, an update to its high-end model focused primarily on autonomous coding tasks and long-running professional work. Benchmark results place the new model ahead of its predecessor and direct competitors in several categories, though behind Claude Mythos Preview, the company's most advanced model.

In agentic coding, Opus 4.7 scores 64.3% on SWE-bench Pro —compared to 53.4% for Opus 4.6 and 57.7% for GPT-5.4— and 87.6% on SWE-bench Verified, above its predecessor's 80.8%. In visual reasoning, it improves from 69.1% to 82.1% without tools, and from 84.7% to 91% with tools. In general knowledge work, the model reaches an Elo score of 1,753 on GDPVal-AA, compared to 1,619 for Opus 4.6 and 1,674 for GPT-5.4.

These vision improvements are partly explained by a concrete technical change: Opus 4.7 accepts images up to 2,576 pixels on the long edge, more than three times the resolution supported by previous versions. Beyond higher resolution, Anthropic notes that the model produces higher-quality results in professional tasks such as creating interfaces, presentations and documents.

On the security front, Opus 4.7 is the first Anthropic model to incorporate automatic safeguards to detect and block requests related to high-risk cybersecurity uses. The company explained that these measures will be tested first on lower-capacity models before being applied to Mythos Preview. Professionals who need to use the model for legitimate tasks such as penetration testing or vulnerability research can apply through the new Cyber Verification Program.

On behavior, Opus 4.7 shows lower rates of deception, sycophancy and cooperation with misuse than Opus 4.6 and Sonnet 4.6, though Mythos Preview remains the best-aligned model according to Anthropic's internal evaluations.

The model has been available since April 16 across all Anthropic products and on platforms including Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Pricing remains the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens.

Key points

• Rises from 53.4% to 64.3% on SWE-bench Pro, surpassing GPT-5.4 (57.7%).
• Visual reasoning improves from 69.1% to 82.1% without tools.
• Accepts images up to 2,576 pixels, more than triple previous versions.
• Higher quality in professional tasks: interfaces, presentations and documents.
• First Anthropic model with automatic cybersecurity safeguards.
• Fewer problematic behaviors than Opus 4.6 and Sonnet 4.6.
• Same price as Opus 4.6: $5 input and $25 output per million tokens.