Claude Opus 5: What's New, Pricing and Who It's For
Claude Opus 5 launched on 24 July 2026 at $5/$25 per million tokens - the same price as Opus 4.8. Here's what actually changed, how it compares to Fable 5 and GPT-5.6 Sol, and which teams should switch.
On this page
Claude Opus 5 launched on 24 July 2026. It costs $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. Anthropic says it comes close to the frontier intelligence of Claude Fable 5 at half the price, and it now leads on coding and knowledge work benchmarks including Frontier-Bench and GDPval-AA.
Most launch coverage quotes the sticker price and moves on. But the rate card has not changed since Opus 4.8, so the price is not the story. What changed is how much you get for each token you spend.
This guide covers what shipped, what it costs in practice, how it compares to the alternatives, and which teams should care.
Claude Opus 5 At A Glance
Opus 5 is the newest model in Anthropic's heavyweight tier. Anthropic positions it as the model you reach for every day, rather than the one you save for special occasions. Fable 5 still leads on the hardest problems, and Mythos 5 remains ahead on cybersecurity and biology work.
- Released: 24 July 2026, available immediately on all platforms.
- API model string: claude-opus-5.
- Price: $5 per million input tokens, $25 per million output tokens.
- Consumer plans: New default model on Claude Max, and the strongest model available on Claude Pro.
- Fast mode: Runs at about 2.5 times default speed, billed at twice the base price. Currently a research preview on the first party Claude API only, so it is not available via Bedrock, Google Cloud or Microsoft Foundry.
- Data retention: No data retention requirements for general access.
The price has not moved. Opus 4.8, which arrived on 28 May, carried the same $5 and $25 rate. You get a much stronger model for the same money, which is not how most of this year's releases have gone.
What's New In Opus 5
Near Fable Performance At Half The Price
On coding and knowledge work evaluations, Opus 5 is the new state of the art. It leads on Frontier-Bench and GDPval-AA, according to Anthropic's announcement. On Frontier-Bench v0.1 it more than doubles Opus 4.8's performance at a lower cost per task.
One caveat worth carrying through this whole section: these are Anthropic's own reported results. The Frontier-Bench figures come from an internal run averaged over five attempts per task, with Opus 4.8 standing in whenever a safety classifier refused a request. Treat them as a vendor's benchmark until independent evaluations land.
The comparisons against Fable 5 are the interesting ones, because Fable costs twice as much:
- CursorBench 3.2: At max effort, Opus 5 lands within 0.5% of Fable 5's peak score at half the cost per task.
- OSWorld 2.0: Opus 5 beats Fable 5's best computer use result at just over a third of the cost.
- ARC-AGI 3: Opus 5 scores three times as high as the next best model on novel problem solving.
- Zapier AutomationBench: About 1.5 times the pass rate of the next best model at the same cost per task. Even at its lowest effort setting, it passes more tasks than any other model.
Anthropic also reports gains across every one of its life sciences evaluations. The largest jumps came in organic chemistry, up 10.2 percentage points on internal benchmarks, and protein sequence work, up 7.7 percentage points.
Effort Scales Better Than On Any Earlier Opus
This is the change with the biggest commercial impact, and it is widely misreported. The effort parameter is not new. It arrived with Opus 4.8 and already works across Sonnet 5, Fable 5, Mythos 5 and Opus 4.5 onwards. It lets you set how much compute the model gives a task, from low through to max.
What is new is how well Opus 5 converts that spend. Anthropic's documentation puts it plainly: Opus 5 turns additional effort into better results more reliably than any earlier Opus model. So the dial is not the story. The exchange rate is.
The context matters here. Fable 5 drew sustained criticism from developers and enterprise buyers over its token burn rate, with teams reporting that they blew through budgets faster than expected. Two early access reports suggest Opus 5 is more frugal:
- A legal AI team held quality at lower reasoning levels while generating 26% fewer tokens on average than Opus 4.8 at max reasoning.
- A trading firm hit its best ever Opus benchmark score using roughly a seventh of the reasoning tokens of Opus 4.8, at under half the latency.
The practical consequence is that a lower effort setting on Opus 5 may now beat a higher setting on Opus 4.8, for less money. So if you are budgeting for Opus 5, test down the effort range rather than defaulting to max. Our cost calculator will help you work out what different token volumes actually cost.
One migration detail for developers: thinking is on by default on Opus 5, where Opus 4.8 left it off unless you set it. Disabling thinking at the xhigh or max effort levels now returns an error, so existing calls may need adjusting.
Better At Checking Its Own Work
Anthropic says Opus 5 is "much stronger at verifying its work and iterating carefully until it succeeds." In practice, that means fewer rounds of back and forth.
Early access testers described the model catching its own logical faults during planning rather than after the fact. One reported it opening its own pages in a browser at desktop and phone widths to check its front end work. Another described it pushing back on a flawed design proposal instead of simply complying.
Two Beta Platform Updates
Anthropic shipped two platform changes alongside the model. Mid conversation tool changes let developers alter which tools Claude can access without invalidating the prompt cache. Automatic fallbacks are opt in, and once enabled they route requests blocked by safety classifiers to another model, so you get an answer instead of an error.
What Opus 5 Costs, And How That Compares
Here is where Opus 5 sits against the models most teams will weigh it against. All figures are per million tokens as of 25 July 2026.
| Model | Input | Output | Positioning |
|---|---|---|---|
| Claude Opus 5 | $5.00 | $25.00 | Everyday frontier work |
| Claude Fable 5 | $10.00 | $50.00 | Hardest autonomous tasks |
| Claude Opus 4.8 | $5.00 | $25.00 | Now superseded |
| GPT-5.6 Sol | $5.00 | $30.00 | OpenAI flagship tier |
Opus 5 Versus Fable 5
Fable 5 is still the stronger model on the hardest problems. Fortune reports that Anthropic still points advanced projects towards Fable 5, including work a model might handle autonomously for days on end. But at twice the price, the question is whether you need that headroom.
For most workloads, the honest answer is no. Opus 5 beats Fable 5 outright on several benchmarks in the launch announcement. Our breakdown of Claude Fable 5 covers where that extra spend still earns its keep.
Opus 5 Versus GPT-5.6
GPT-5.6 Sol, released on 9 July, matches Opus 5 on input at $5 but charges $30 on output against Opus 5's $25. OpenAI also pushed economical token use as a selling point for that release, so both companies now compete on cost efficiency rather than raw capability alone.
One detail to watch: GPT-5.6 Sol pricing rises to $10 and $45 for long context requests above roughly 272K input tokens. See our wider ChatGPT vs Claude vs Gemini comparison for the full picture.
Why Cost Per Task Beats Cost Per Token
Rate cards are a poor guide to what you will spend. A model that costs less per token but needs three attempts to get something right is more expensive than one that costs more and succeeds first time.
This is why Anthropic reported its Frontier-Bench and CursorBench results as cost per task rather than cost per token. It is the number you should track too. If you are unsure how token billing works, start with our explainer on what a token actually is.
Safety, Guardrails And Data Retention
Anthropic describes Opus 5 as its most aligned model to date. It scores 2.3 on the company's automated behavioural audit for overall misaligned behaviour, the lowest of its recent releases. Anthropic also reports the lowest rates of deceptive behaviour and the least susceptibility to being tricked into misuse.
On capability risk, Opus 5 sits behind Mythos 5 in both biology research and offensive cybersecurity. On the cyber side Anthropic frames this as deliberate, saying it has intentionally avoided training Opus 5 on cyber tasks. The practical detail for developers is that cyber classifiers should trigger roughly 85% less often than they do on Fable 5.
Opus 5 can find vulnerabilities in source code. It is blocked from binary based vulnerability scanning, penetration testing and exploit generation.
Privacy conscious buyers should note one more thing. Like Opus 4.8, and unlike Fable and Mythos, Opus 5 is not subject to the 30 day data retention policy. For teams in regulated sectors, that alone can decide the choice. TechCrunch covers the guardrail changes in more depth.
Who Claude Opus 5 Is For
Developers And Engineering Teams
This is the clearest case. Opus 5 leads on agentic coding benchmarks and holds context across long multi step work.
Lovable reported a 22% improvement over Opus 4.7 on its hardest agentic coding tasks, with far less variation between runs. For production pipelines, that consistency often matters more than a peak score.
Business And Knowledge Work Teams
The AutomationBench results are the signal here, and Zapier's own account of testing it is the useful detail. Zapier reports that Opus 5 took a raw account health workbook and ran a full churn prevention sequence end to end: flagging at risk accounts, alerting the right owner and summarising for retention ops. Previous models did not pass. Opus 5 scored 100%.
If your use case is document analysis, due diligence, financial modelling or report generation, this is a meaningful upgrade rather than an incremental one.
Researchers And Scientists
Opus 5 is now Anthropic's most capable generally available model for scientific research. Biology requests previously blocked on Fable 5 will route to Opus 5 rather than Opus 4.8. For long running autonomous research, Mythos 5 remains stronger.
When To Still Choose Fable 5
Three situations still justify the higher rate:
- Multi day autonomous work: Tasks a model will run unsupervised for days at a time.
- Absolute frontier problems: Work at the limit of what any current model can do.
- Restricted capabilities: Work that needs capabilities Opus 5's guardrails block.
For everything else, Opus 5 is likely the better economic choice.
Four Frontier Models In Under Two Months
Opus 5 is Anthropic's fourth 5 series release in less than two months, following Mythos 5, Fable 5 and Sonnet 5 in June. Opus 4.8 landed shortly before that run, on 28 May. Only Haiku still awaits its 5 series upgrade.
If you manage AI spend, that cadence is the real story. A model you benchmarked and priced in May may already be two generations behind. Our July 2026 AI news roundup tracks how quickly the landscape has shifted, and our pricing history charts show the longer trend.
Fun Fact: The Model That Built Its Own Eyes
On one Frontier-Bench task, Opus 5 was handed a drawing of a machine part and asked to write code that would rebuild it as a 3D FreeCAD model. The catch: it was deliberately given no way to view the drawing. Rather than giving up, it wrote its own computer vision pipeline to pull the geometry out of the raw pixels, then reconstructed the full part. It succeeded repeatedly. Anthropic reports that no competing model with the same setup solved it in five attempts.
Should You Switch To Opus 5?
If you are on Opus 4.8, the answer is almost certainly yes. You get the same price, much better performance, improved token efficiency and stronger alignment. There is no cost argument for staying.
If you are on Fable 5, run your own evaluation before you assume you need to stay. Halving your rate while losing little measurable capability is a rare trade to be offered.
If you are on GPT-5.6 or another provider, the comparison is closer. It depends on your workload, your existing tooling and how much the data retention difference matters to you. Model your real token volumes rather than comparing rate cards.
You can compare Opus 5 against every model in Anthropic's full pricing lineup and nine other providers on our homepage. Work out what a switch would save you with the pricing calculator, or plan a multi model setup with the AI API comparison. If you decide to move, our migration guides include working code examples.
Frequently Asked Questions
How Much Does Claude Opus 5 Cost?
Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. Fast mode is billed at twice the base price. Full details are in our Claude pricing FAQ.
When Was Claude Opus 5 Released?
Anthropic released Claude Opus 5 on 24 July 2026. It was available immediately across all platforms and paid plans.
Is Opus 5 Better Than Fable 5?
On several benchmarks in Anthropic's launch announcement, yes. These include Frontier-Bench, ARC-AGI 3 and OSWorld 2.0. Those are vendor reported figures, so treat them accordingly. Fable 5 remains stronger on the hardest problems and on long running autonomous work, but Opus 5 delivers close to Fable level performance at half the price.
How Is Opus 5 Different From Opus 4.8?
Opus 5 more than doubles Opus 4.8's Frontier-Bench v0.1 performance at a lower cost per task, on Anthropic's own internal run. It improves on every life sciences evaluation, generates fewer tokens for the same quality of output, and scores as Anthropic's most aligned model to date. Thinking is also on by default, where Opus 4.8 left it off. The price is unchanged.
Is The Effort Setting New In Opus 5?
No. The effort parameter arrived with Opus 4.8 and works across Opus 4.5 onwards, Sonnet 5, Fable 5 and Mythos 5. It controls how much compute the model gives a task, from low up to max. What is new is that Opus 5 converts extra effort into better results more reliably than any earlier Opus model, so the same spend buys more.
The Bottom Line
Opus 5 is the clearest value improvement Anthropic has shipped this year. You get the same price as its predecessor, close to Fable 5 capability at half the cost, and better returns on every token you spend on reasoning.
The wider lesson is that frontier pricing now moves fast enough that periodic review is not optional. Keep an eye on the numbers at compareai.today, where we track live API pricing across ten providers.
