Anthropic: Claude Opus 4.7 released with stronger coding and higher-resolution vision

Anthropic released Claude Opus 4.7 on April 16, 2026. The model is generally available across Claude.ai, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, at the same pricing as its predecessor: $5 per million input tokens and $25 per million output tokens. Developers access it via the model ID claude-opus-4-7.

The most significant improvement for product teams running agentic workflows is coding reliability. The model catches logical faults during the planning phase rather than mid-execution, and produces fewer errors when calling tools in multi-step tasks. For teams already using Claude in automated pipelines, this reduces the frequency of manual corrections during long-running operations.

Vision capabilities have expanded considerably. Opus 4.7 accepts images up to 2,576 pixels on the long edge — roughly 3.75 megapixels, more than three times the resolution supported by prior Claude models. This makes the model more practical for tasks like reading dense UI screenshots, interpreting technical diagrams, or processing design files within agent workflows.

A new task budgets feature gives the model a token estimate for a full agentic loop. Claude Opus 4.7 uses this countdown to prioritise work and complete tasks within the allocated budget. For product teams running production agents, this produces more predictable compute costs and more graceful task completion when resources run low.

The model uses a new tokenizer, which may increase token consumption by 1x to 1.35x compared to previous versions. Teams estimating API costs for existing prompts should account for this when migrating.

Why it matters for product managers

For PMs building AI features that rely on Claude, Opus 4.7 is most relevant if your current implementation involves coding agents, complex document analysis, or image-intensive workflows where prior versions produced unreliable results. The improved instruction following is also worth noting: prompts calibrated for earlier models may need retuning, as the model interprets instructions more precisely than before.