Opus 4.7 Token Increase

Claude Opus 4.7 has been officially released. Anthropic stated that this model is stronger than Opus 4.6 in high-difficulty software engineering, long-running tasks, instruction following, and self-verification.

The core change is the updated tokenizer. Text processing has improved, but the same input may map to about 1.0-1.35x more tokens depending on content type, which may require readjusting existing prompts and harnesses.

The key improvements are as follows:

• High-resolution vision: supports images up to 2,576px on the long side
• Business output: interfaces, slides, and document quality have improved
• Efficiency: Opus 4.7 at low effort settings generally approaches the level of Opus 4.6 at medium effort
• Benchmark: +13% over Opus 4.6 on a 93-task coding benchmark
• Deployment: available on Claude products, API, Amazon Bedrock, Vertex AI, and Microsoft Foundry
• Pricing: input $5/M, output $25/M, the same as 4.6

On the security side, guardrails have been added to automatically detect and block high-risk cyber requests, and users with legitimate security purposes can participate in the Cyber Verification Program. Overall safety is similar to 4.6, though some metrics have improved while others have slightly weakened.