Claude Opus 5 in Cursor: Pricing, Context and When to Use It

Claude Opus 5 is Anthropic's latest Opus model and the direct replacement for Opus 4.8 in Cursor's model picker. Cursor describes it as a step-change over Opus 4.8 on coding with the agent, professional knowledge work and long-horizon reasoning, and says it lands on par with Fable 5 on CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. while charging Opus pricing. Cursor's own recommendation is to run the high thinking variant for the best results.

The model card is the fastest way to see where it sits against the rest of the picker, so start there before the pricing detail.

Cursor publishes both sides for this model, which makes the selection decision unusually concrete. Its strengths cluster around sustained, multi-step work, the kind where a model has to hold a goal across a long session and adapt when a tool returns something it did not expect. The two limitations it names are cost and verbosity.

Read the second row as a scheduling argument, not a pricing one. If Opus 5 and Fable 5 land in the same place on hard coding work but Opus 5 gets there sooner at Opus rates, the saving shows up in elapsed time and in the tokens a longer session would have burned. The headline rate stays where it is.

The verbosity limitation shows up in review time. A one-file fix comes back with the fix plus two adjacent tidy-ups nobody asked for, and since none of it is actually wrong, none of it can be rejected on sight. The whole diff gets read before any of it is approved.

Yes, and Cursor calls this out as a difference from Fable 5, not as a general property of the Claude family. For teams whose model allow-list comes out of a security review, this is usually the line that settles the choice.

Opus 5 draws from the third-party Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool, which is metered separately from the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool. Other Models access starts with Pro at $20 of monthly usage, with more on Pro Plus and Ultra. Cursor StartAn India-only individual plan centered on Cursor Models, with cloud and mobile access but without Other Models, Auto, Bugbot, Automations or the Agent SDK. Press Enter for the full definition. includes only Cursor Models, so it does not include Opus 5. All rates below are per million tokens.

The Fast tier (claude-opus-5-fast) was available at launch and exists for higher-priority output, at double the standard input and output rates. On legacy request-based plans it requires Max Mode. Cursor's guidance is to use it selectively, on time-sensitive or critical work.

Set the included $20 against the output rate and it stops looking generous. At $25 per million, $20 of Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. usage is 800,000 output tokens if nothing else draws on it, and Cursor's own sizing guidance puts daily agent users at $60 to $100 a month in total usage. Pro Plus includes $70 and Ultra $400. For a solo developer on Pro, the included pool covers trying Opus 5 rather than working in it all month.

No. Opus 5 bills at the same per-token rates however large the context gets. That is not how every model in the picker works, so a pricing habit built on another one may not carry over.

Opus 5 also supports a thinking variant for deeper reasoning, and Cursor recommends the high setting for the strongest results. That puts the cost question on thinking tokens at $25 per million output. A large repository dump does not trip a penalty rate on this model, but high-effort reasoning still bills every token it produces.

The habit the no-multiplier rule breaks is watching the token count to stay the right side of a threshold. On Opus 5 there is no higher rate waiting on the other side of it, so the watching buys nothing. Compaction still pays for itself, though, for a reason that has nothing to do with the bill. A Cursor field engineer's rule of thumb puts a slight drop in how well a model performs at roughly 50 to 60% of a filled window, so /summarize or a fresh thread stays the right move on a long Opus 5 session.

Four things move Opus 5 spend off the published $5 and $25. The Fast tier doubles both, to $10 input and $50 output. On Teams and Enterprise plans, third-party model requests also carry the Cursor Token Rate of $0.25 per million tokens. Opting in to regional data residency adds a 10% uplift on model pricing for eligible models. And on legacy request-based plans, where the Fast tier requires Max Mode, Max Mode itself bills at the model's API rate plus 20%.

Check them in that order. Or rather, check which pool the spend is landing in before any of the four. Both pools show in editor settings and on the usage dashboard, and a bill dominated by the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool is not an Opus 5 question at all.

Once the spend really is landing in Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition., the Fast tier and the Cursor Token Rate are the two I would expect to get misread, since a doubled rate and a fee that rides on every third-party token both look like the model simply being expensive. A team reading it that way starts rationing thinking depth when the cause was that it had been running claude-opus-5-fast all along.

Residency is enrolled at the organisation level rather than chosen per request, which is what makes it easy to miss when you are reconciling a bill.

Cursor's tiering puts Opus 5 at high cost and medium speed, which sets the boundary. It earns its rate on work that would otherwise take several attempts with a cheaper model. Where that line falls is not a number Cursor publishes, and it depends on the codebase as much as on the model. The pairings below come straight out of Cursor's published strengths and limitations.

On Teams and Enterprise plans there is one more option worth weighing. Instead of pinning Opus 5 for everyone, Cursor Router can send only the hard requests to a model at this tier and keep routine work off it. That is the same argument the router's launch data makes, measured against pinning everything to Opus 4.8.

Which lever you have depends on the plan, though. Cursor Router is a Teams and Enterprise feature, so it is not a lever you have on an individual plan, and what decides the monthly bill is whether you drop back to Sonnet 5 or Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. once the hard part of a task is done. Cursor's own figure is that roughly 60% of developers settle on one model as their daily driver. Check whether yours is Opus 5 before deciding the rates are the problem.