Claude Opus (currently Opus 5.5) is Anthropic's model for long-running agentic coding and knowledge work: a 1M-token context window with more reasoning depth than Sonnet, at $4/$20 per million input/output tokens. Pick it when a task genuinely needs that depth over many steps; use Sonnet for everyday work and save Opus for what actually needs it.
Other good fits
Agentic work that runs for many tool calls in a row, where reasoning quality compounds over the whole run.
+ Pros
Read pros
Opus is Anthropic's model for demanding, long-horizon work: Anthropic describes the current version, Opus 5.5, as built "for long-running agentic coding and knowledge work." It shares Sonnet's 1M-token context window but is positioned above it for tasks where reasoning depth, not just speed, decides whether the result is right.
Pick it when…
A task runs over many steps (an agent making dozens of tool calls, a multi-file refactor) where a weaker model's small mistakes compound.
You've tried Sonnet at a higher reasoning effort and it still falls short on the specific task.
The cost of a wrong answer is high enough that paying more per token for better reasoning is clearly worth it.
− Cons
Read cons
Opus costs twice Sonnet's per-token rate for input and output, so using it as a default for routine work is an easy way to overspend without a matching improvement in results. Like every Claude model, it's closed and cloud-only, so anything that has to run offline or on your own hardware needs a different kind of model entirely.
Pick something else when…
The task is routine, not demanding.Claude Sonnet handles everyday coding and writing for less than half the price.
You're doing simple, high-volume work.Claude Haiku is faster and cheaper for classification, extraction and routing.
Even Opus falls short on a specific hard task. Anthropic recommends Claude Fable for demanding reasoning and long-horizon agentic work beyond what Opus handles.
Avoid if
You're doing simple, high-volume classification; [Claude Haiku](tool page) is built and priced for that instead.
Context
Opus 5.5 has a 1M-token context window and a 128K-token maximum output on the synchronous API (up to 300K on the batch API with a beta header). Its reliable knowledge cutoff is June 2026. Anthropic commits to keeping it available on its own platforms no sooner than 22 September 2027.
Requirements
Opus is a closed, cloud-hosted model with no self-hosting or offline option. Use it through the Claude API, claude.ai (Pro and above), or a supported cloud platform (Amazon Bedrock, Google Cloud, Microsoft Foundry), each with its own lifecycle dates.
Licence & cost
Prices checked 23 September 2026 (pricing (opens in a new tab)): $4 per million input tokens, $20 per million output tokens, through the Claude API. Batch requests run at 50% off; prompt-cache reads cost about 5% of the base input price. It's also available through Claude Pro and Max on claude.ai as one of the "additional models." A long agentic run at Opus's price can add up quickly, so it's worth confirming the task actually needs Opus before defaulting to it.
First task
Documentation-based suggestion — not run here.
Give it a task that actually needs multiple steps and judgment calls, for example planning and then implementing a small but genuinely multi-file change to a codebase, and compare the result against running the same task on Sonnet. Look specifically for places where Opus caught an edge case or dependency that Sonnet missed.
Done when you can point to at least one concrete difference in the outcome, not just a subjective sense that the answer "felt" better.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Anthropic's models overview and pricing pages, read on 2026-09-23. No model calls were made for this review; this describes the current Opus generation (Opus 5.5) and will need updating when Anthropic ships the next one.
Claude models overview · Sep 23, 2026 — Opus 5.5 specs: 1M context, 128K max output, June 2026 knowledge cutoff, retirement not before September 2027.
Claude API pricing · Sep 23, 2026 — Opus 5.5 at $4/$20 per million input/output tokens; batch and cache-read discounts.
Choosing a model · Sep 23, 2026 — Guidance on when Opus is the right choice over Sonnet or Fable.
Further review notes & alternatives
Alternatives and trade-offs
Claude Sonnet wins on cost for anything that doesn't need Opus's extra reasoning depth.
Claude Haiku wins for simple, high-volume tasks where speed and price matter more than depth.
GPT-5 wins as the closest cross-vendor comparison for demanding reasoning work; worth benchmarking both on your actual task.
Use a large-context model for demanding coding, analysis, writing, and agentic tasks.
Choose it when long, multi-step work needs adjustable reasoning effort.
It accepts images and provides a 1M-token context window.
