Claude Opus 5 is Anthropic's Opus-tier model, released July 24, 2026. It costs $5 per million input tokens and $25 per million output, unchanged from Opus 4.8, and Anthropic positions it as close to Claude Fable 5 in capability at a fraction of the price. It is now the default model on Claude Max, with a 1 million token context window and thinking on by default.
What is Claude Opus 5?
It is the successor to Opus 4.8, and the second model in Anthropic's Claude 5 generation after Claude Fable 5. The API id is claude-opus-5.
The important thing to understand about where it sits: Opus 5 is not the top of the Claude lineup. Fable 5 and the restricted Mythos 5 still occupy the Mythos class above it. What makes Opus 5 interesting is the gap it closes from below. Anthropic's own launch material is built almost entirely around price-performance comparisons rather than around a new capability ceiling, which is a first for a Claude release covered on this site.
What does Claude Opus 5 cost?
$5 per million input tokens and $25 per million output tokens, the same as Opus 4.8 before it. There is also a fast mode that runs at 2.5 times the default speed for twice the base cost.
For comparison, Fable 5 costs $10 per million input and $50 per million output. So Opus 5 is half the price of the model directly above it, and the same price as the model it replaces. If you were already budgeting for Opus, your costs do not change.
How does Claude Opus 5 compare to Fable 5?
By Anthropic's own numbers, close on capability and much cheaper. Every figure below is vendor-reported, gathered by Anthropic on its own evaluations, and no independent measurement existed at the time of writing:
- On CursorBench 3.2, within 0.5 percent of Fable 5's peak score, at half the cost.
- On OSWorld 2.0, ahead of Fable 5, at one third of the cost.
- On Frontier-Bench v0.1, above all competitors and double Opus 4.8's score at lower cost.
- On ARC-AGI 3, three times the score of the next-best model.
- On Zapier's AutomationBench, roughly 1.5 times the pass rate of the next-best model at equivalent cost.
Cursor's co-founder summarized it at launch as near Fable 5 intelligence at Opus speed and cost. Lovable's co-founder called it the biggest leap in the Opus family since 4.5.
The honest reading is that Anthropic is making a value argument, not a frontier argument. On scientific work it reports Opus 5 at 10.2 percentage points above Opus 4.8 on organic chemistry tasks and 7.7 points above on protein prediction. Those are real gains over its predecessor. They are not claims to have passed the Mythos-class models, and on cybersecurity exploitation specifically Anthropic states Opus 5 remains behind Mythos 5.
Launch benchmarks are also the least reliable moment to judge a model. Community rankings accumulate over weeks, and the Claude benchmarks page tracks where the family actually lands as weekly LMArena snapshots come in.
Where can you use Claude Opus 5?
Everywhere Anthropic ships, from launch day: the Claude API on all cloud platforms, Claude.ai, Claude Code, Claude Cowork, and the Claude Pro and Claude Max subscriptions. On Max it is now the default model, which means a large number of people are using it whether or not they went looking for it.
One detail worth knowing if you are building on the API: Opus 4.8 stays available as a fallback for requests that trip a safety classifier. Anthropic reports that Opus 5's cyber classifiers intervene about 85 percent less often than Fable 5's.
That fallback matters for a reason beyond safety. If a session is quietly handed to a different model partway through, the experience changes without any announcement, and that is one of the ordinary explanations behind a sudden sense that quality dropped. Our guide to whether Claude is getting dumber covers how to tell a real regression from a bad day.
What is different about using Opus 5 day to day?
Its default behavior changed more than its price did. Anthropic's own documentation describes a model that talks more, narrates its work more, verifies itself without being asked, delegates to subagents readily, and can widen the scope of a task you meant to keep narrow. Several instructions that were good practice on older models now work against it, and Anthropic reports removing more than 80 percent of Claude Code's system prompt for this generation with no measurable loss on its coding evaluations.
If you are carrying prompts, project instruction files, or harness scaffolding forward from an earlier model, that is the part worth your attention. We cover it in detail in prompting Claude 5 models.
How are people rating Claude Opus 5?
Too early to say, and anyone claiming otherwise a day after launch is guessing. Two things reliably distort the first weeks of any model release: the honeymoon effect, where early impressions run high because expectations and attention are both elevated, and the fact that the people testing a model on day one are not a representative sample of the people who will use it in a month.
This site measures the reaction as it happens rather than predicting it. The live Claude tracker records how users rate their sessions on a five-point scale, day by day, with optional version tags and short written reports attached to votes. The daily record from the July 24 launch onward is preserved in the reviews archive. Sample sizes on any single day are small, so read a single day as an anecdote and a trend across weeks as a signal; how votes are collected and what they can and cannot show is documented on the about page.
Other releases in the family are covered in the Claude models section.
FAQ
Is Claude Opus 5 better than Claude Fable 5?
Not on raw capability, by Anthropic's own framing. Fable 5 sits in the Mythos class above Opus and remains the more capable tier for long-horizon and ambiguous work. Opus 5's claim is that it gets close on many tasks for half the price, and beats Fable 5 on some agentic benchmarks at a third of the cost. For most everyday coding, that trade favors Opus 5.
Does Claude Opus 5 cost more than Opus 4.8?
No. Pricing is identical at $5 per million input tokens and $25 per million output tokens. The optional fast mode, which runs at 2.5 times the speed, costs twice the base rate.
What is the context window on Claude Opus 5?
1 million tokens, as both the default and the maximum. Anthropic says its instruction following and reasoning stay consistent across that whole window. Treat the advertised number and the usable number as different things: in practice, quality of decisions still degrades well before a window that size is full, which is why our guidance on scaling AI development recommends budgeting around 100,000 tokens per session regardless of what the model advertises.
Should I switch from Opus 4.8 to Opus 5?
Anthropic says Opus 5 performs well out of the box on existing Opus 4.8 prompts, and the price is the same, so there is little reason not to try it. Two API changes to know about: thinking is on by default, and it can only be disabled at high effort or below. Verification and re-check instructions written for older models are worth deleting rather than carrying over.
Is Claude Opus 5 available on the free plan?
Anthropic lists availability on Claude Pro and Claude Max, where it is now the default model on Max, alongside the API, Claude.ai, Claude Code, and Claude Cowork. Free-tier access was not part of the launch announcement.