GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone

Claude Opus 5 — 1M Context, Agentic Coding

Claude Opus 5 is Anthropic's model for complex agentic coding and enterprise work. It gives you a 1M-token context window, up to 128k output tokens per call with thinking tokens counted inside that budget, adaptive thinking that is on by default, and a five-level effort dial — low, medium, high, xhigh, max — that takes the place of the sampling knobs, since temperature, top_p and top_k are rejected on this model.

This page introduces the model. API access for this model is not currently available on reAPI. Browse the model catalog for available services.

Highlights
Extended thinking · tool use · structured output
Best for
Hard reasoning, agent workflows
Input
Chat messages (OpenAI-compatible)
Output
Text, streaming optional

What is the Claude Opus 5 API?

Claude Opus 5 is Anthropic's model for complex agentic coding and enterprise work. It gives you a 1M-token context window, up to 128k output tokens per call with thinking tokens counted inside that budget, adaptive thinking that is on by default, and a five-level effort dial — low, medium, high, xhigh, max — that takes the place of the sampling knobs, since temperature, top_p and top_k are rejected on this model.

What you can build with this model

Real-world workflows and production use cases you can build and ship with this model.

Claude Opus 5 working through a multi-file code change in one session

Long-horizon agentic coding that finishes the job

Anthropic positions Claude Opus 5 for complex agentic coding and enterprise work, and the shape of the model matches that: a 1M-token context window means a whole repository, its tests and its issue history fit in one request, and 128k output tokens is room for a multi-file change rather than a stub. Adaptive thinking is on by default, so Claude Opus 5 holds a plan together across tool calls instead of restarting its reasoning each turn. The practical advice from Anthropic's own migration guidance is to give Claude Opus 5 the complete task specification up front in a single well-specified turn and let it run, rather than revealing the task across many short interactive turns.

Claude Opus 5 reviewing a pull request and reporting findings with severity

Code review and debugging with the filter turned off

Claude Opus 5 finds real bugs at high precision and high recall, and it stays accurate at lower effort — which makes a cheap fast pass at review time plus a thorough pass later a practical pattern. One caveat worth designing around: Claude Opus 5 follows severity instructions literally. A review prompt that says only report high-severity issues will get exactly that, and your measured recall drops even though the underlying bug-finding improved. Ask Claude Opus 5 to report everything with a confidence level and a severity, then filter in a separate pass.

Claude Opus 5 reading a large document set and producing a structured deliverable

Whole-document enterprise work without a chunking layer

Text, images and PDF are all inputs to Claude Opus 5, and the million-token window is large enough that a contract set, a research bundle or a year of filings goes in whole. That removes the retrieval layer whose only job was working around a small context. Claude Opus 5 also generates and edits complex multi-sheet spreadsheets with real formulas and builds slide decks that follow design conventions, and it can be told to follow a specific template when one is required. Vision is strong enough on charts and dense documents that giving Claude Opus 5 tools to crop and re-examine its own output beats simply raising the thinking depth.

Claude Opus 5 vs Claude Opus 4.8

This comparison is not about price. Anthropic publishes the same per-token rate, the same one-million-token context window and the same 128k output ceiling for both models. Claude Opus 5 is the capability step-change on top of that, and it quietly changes two request-surface defaults that will break code carried straight over — thinking now runs by default, and turning it off is capped at high effort.

Capability
Claude Opus 5
Claude Opus 4.8
Anthropic's positioning
For complex agentic coding and enterprise work; the current Opus generation
The most capable model of the Opus 4 series, now the previous generation
Context and output limits
1M-token context, 128k max output
Identical — 1M-token context, 128k max output
Thinking default
On — omitting the thinking parameter runs adaptive thinking
Off — adaptive thinking has to be set explicitly
Disabling thinking
Accepted only at effort high or lower; rejected at xhigh and max
Accepted at any effort level
Prompt-cache minimum
512-token minimum cacheable prefix, so shorter prompts now cache
1024-token minimum — twice as long a prefix before caching starts

Comparison reflects behavior documented on Anthropic's own model-overview page at the time of writing.

Frequently asked questions

Common questions about this model.

This page introduces the model. API access for this model is not currently available on reAPI. Browse the model catalog for available services.

Claude Opus 5

Claude Opus 5

This page introduces the model. API access for this model is not currently available on reAPI. Browse the model catalog for available services.