ClaudeUpdated Apr 18, 20268 min

Claude Opus 4.7 is official: new coding autonomy, xhigh effort, and token budgets

Anthropic officially launched Claude Opus 4.7 on April 16, 2026. The release matters for builders because it directly targets long-running software engineering work while also changing the token economics of Claude Code and API-based agent workflows.

What happenedOpus 4.7 is generally available across Claude products, the API, Bedrock, Vertex AI, and Microsoft Foundry.

Builder signalAnthropic highlights harder coding tasks, better instruction following, higher-resolution vision, and file-system memory.

TRH actionRetune prompts, measure token deltas, and set explicit effort and task budgets before upgrading production agents.

What Anthropic confirmed

Anthropic says Opus 4.7 is a direct upgrade from Opus 4.6, with stronger performance on difficult software engineering tasks, complex long-running work, precise instruction following, and self-verification before reporting results. It also supports higher-resolution image understanding, up to 2,576 pixels on the long edge, and better use of file-system based memory across multi-session work.

The model name for developers is claude-opus-4-7. Anthropic says pricing remains the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens.

The Claude Code updates are the real operational story

The release adds a new xhigh effort level between high and max. Anthropic also raised Claude Code's default effort level to xhigh for all plans and recommends starting with high or xhigh for coding and agentic use cases. Claude Code also gets /ultrareview, a dedicated review session intended to find bugs and design issues, plus expanded auto mode for Max users.

For builders, that is not just a capability update. It is a budget update. Higher effort can improve reliability on hard tasks, but it can also increase output tokens and latency if the task is underspecified.

Token usage changed in two ways

Anthropic explicitly warns that Opus 4.7 uses an updated tokenizer, so the same input can map to roughly 1.0 to 1.35 times as many tokens depending on content type. The model can also think more at higher effort levels, especially on later turns in agentic settings. Anthropic's own migration guidance is to measure real traffic, control effort, use task budgets, and prompt for concision where appropriate.

What Token Robin Hood readers should do

Treat Opus 4.7 as an upgrade that needs an operating harness, not a simple model swap. Pin the model in critical workflows, run a before-and-after token audit, cap effort by task type, downsample images when extra visual detail is unnecessary, and separate planning, execution, and review into measurable phases. The teams that win from Opus 4.7 will be the teams that convert extra autonomy into shipped artifacts rather than invisible token burn.