Claude Opus 5 Pricing: $5 and $25 per Million Tokens

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, the same rate as Opus 4.8 and exactly half of Claude Fable 5's $10/$50 pricing. Anthropic released the model on July 24, 2026, and priced it to undercut its own flagship rather than sit above it.

One update before the rest: on September 22, 2026 Anthropic released Claude Opus 5.5 at $4/$20, and Opus 5 is now listed as a legacy model. Opus 5 is still available at the same $5/$25, so everything below about Opus 5 still holds. The next section covers what changed.

Prices checked September 26, 2026 and re-checked October 24, 2026 against Anthropic's pricing page and models overview.

Opus 5.5 (Sept 22, 2026): $4/$20

Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. That is 20% below Opus 5 on both input and output. Anthropic's models overview now names Opus 5.5 as the model to start with for most workloads, and its Opus 5.5 model page gives the release date as September 22, 2026.

Cache writes follow the same pattern. The 5-minute and 1-hour tiers cost $5 and $8 per million tokens on Opus 5.5, against $6.25 and $10 on Opus 5. Cache reads drop further than the base rate. Anthropic prices Opus 5.5 cache hits at 5% of base input ($0.20), where Opus 5 uses the standard 10% ($0.50).

Which one to use. Start new work on Opus 5.5. Move up to Fable 5.1 only when your own evals on Opus 5.5 at higher effort still fall short, which is the rule Anthropic's models overview gives. Drop to Sonnet 5 when speed or cost matters more and it passes your evals. Since September 28, 2026 the current Sonnet is Claude Sonnet 5.5, at the same $2 and $10 with cache reads halved to $0.10, though five breaking changes mean moving to it is more than a model ID swap. Stay on Opus 5 for now only if your code depends on something Opus 5.5 changed: Anthropic lists four breaking changes, including thinking that can't be disabled and forced tool use returning an error. Plan the migration rather than staying put.

Our Claude Opus 5.5 release breakdown covers the breaking changes in detail.

Claude Opus 5 pricing at a glance

Opus 5 also has a batch rate of $2.50/$12.50 and cache reads at $0.50 per million tokens, per Anthropic's pricing page.

On price specifically, Opus 5 lands at the same $5/$25 rate as Opus 4.8, not above it. If you were already budgeting for Opus 4.8, nothing changes on the invoice. If you were paying Fable 5 rates for work that Opus 5 now handles as well or better, your cost per million tokens is cut in half.

Why Anthropic is talking about cost per task, not cost per token

Token price is easy to compare and easy to publish, so it is what gets quoted first. It is also an incomplete measure of what a model actually costs you to run.

Anthropic's own benchmark disclosures for Opus 5 make the cost-per-task framing explicit. On OSWorld 2.0, the computer-use benchmark, Anthropic says Opus 5 beats Fable 5 at roughly one third of the cost. That is not a token-price claim. It is a claim about how many steps, retries, and tokens a model needs to actually finish a computer-use task end to end.

Here is why that distinction matters for a real workload. A model priced at half the tokens can still cost you more per completed job if it takes twice as many attempts, generates longer chains of reasoning, or needs a second pass to fix its own mistakes. Conversely, a model with a higher per-token price can be cheaper in practice if it finishes the task in fewer steps. Anthropic is arguing Opus 5 wins on both axes for computer-use workloads: cheaper tokens and fewer of them needed per task.

For a straight text-generation or coding workload where token count scales predictably with output length, the $5/$25 rate is a direct, comparable number. For agentic workloads, especially ones that involve browsing, tool calls, or multi-step computer control, the cost-per-task framing is the one to actually budget against.

How Opus 5 pricing compares to Claude Fable 5

If you are deciding between Opus 5 and Fable 5 for a new project, the pricing gap is the first filter, not the last one.

• Opus 5 at $5/$25 per million tokens was the model to default to for the large majority of workloads at its launch: coding, writing, analysis, agent orchestration, and computer-use tasks where Opus 5 is near Fable 5 performance at half the price, and ahead of Fable 5 on OSWorld 2.0. Since September 22, 2026, Anthropic points new work to Opus 5.5 instead, which is cheaper still at $4/$20.
• Fable 5 at $10/$50 per million tokens remains double the price. There is no benchmark disclosed where Fable 5 shows a decisive win over Opus 5 on general intelligence; on the Artificial Analysis Intelligence Index v4.3.2, Opus 5 scores 51 against Fable 5's 50 (#11 of 211, checked September 26, 2026).

That combination, near Fable 5 performance at half the price, and ahead of Fable 5 on OSWorld 2.0, is why most teams currently on Fable 5 for non-specialized work have a straightforward case to test Opus 5 against their own evals before renewing budget at the higher rate. For a deeper look at where Fable 5 still makes sense, see our Claude Fable 5 pricing breakdown and the current Claude Fable 5 alternatives if you are shopping for something other than Opus 5 entirely.

Who should actually switch to Opus 5

Not every team benefits from switching models the day a new one ships. Here is a practical way to think about it. One caveat added September 26, 2026: if you are choosing a model today rather than moving off Fable 5, start with Opus 5.5 (see the Opus 5.5 section above). The lists below still apply to Opus 5.

Switch now if:

• You are running Fable 5 for coding, research, writing, or general agent tasks where Anthropic has not claimed a Fable 5 advantage.
• You run computer-use or browser-automation agents, where the OSWorld 2.0 cost claim (roughly a third of the cost at a win) is the strongest financial argument in this release.
• You use Zapier or similar automation tooling; Anthropic cites Opus 5 at roughly 1.5x the next-best score on Zapier's AutomationBench, a workload category priced identically to any other Opus 5 usage. Anthropic's September table shows Opus 5 at 26.9%, behind Opus 5.5, Fable 5.1, GPT-6 Astra and GPT-5.6 Sol.

Stay on your current model if:

• Your workload is cybersecurity-specific. Anthropic's own disclosure places Opus 5 behind Mythos 5 on cybersecurity benchmarks, the one category in this release where Anthropic did not claim the top spot.
• You are already on Opus 4.8 and your evals show no material gap versus Opus 5 on your specific task. The token price is identical, so there is no cost reason to move until you have benchmarked the quality difference for your use case.
• You have hard-coded model version pins in a regulated pipeline where a mid-quarter model swap requires a change-control cycle regardless of price.

What this means for your monthly bill

A rough way to estimate the shift: take your current Fable 5 monthly spend, divide by two, and that is your Opus 5 spend for the same token volume, assuming equivalent output length. In practice, output length is not always equivalent, since Opus 5's disclosed Frontier-Bench performance (which Anthropic says beats all other models tested, including more than 2x Opus 4.8) suggests it may need fewer corrective turns on complex tasks, which would push real savings above the raw 50% token-price cut.

The same token-only math with Opus 5.5 in the picture: Opus 5.5 spend is 0.8 times Opus 5 spend and 0.4 times Fable 5 or Fable 5.1 spend for the same token volume, before cache reads. Anthropic's pricing page uses a one-hour coding session of 50,000 input and 15,000 output tokens as its worked example. Recomputed on token cost alone:

These figures leave out anything billed outside tokens (Anthropic's own example adds $0.08 of Managed Agents session runtime) and assume each model uses the same number of tokens for the task, which is rarely exact.

The only way to know your actual number is to run your own workload through both models and compare total tokens consumed to completion, not just the per-token rate. For a side-by-side on where Opus 5 and Fable 5 actually diverge on capability, not just price, see our full Opus 5 vs Fable 5 comparison. For the benchmark detail behind these pricing decisions, read our Claude Opus 5 benchmarks breakdown. You can also browse current pricing and specs for every Anthropic model on our models page.

FAQ

How much does Claude Opus 5 cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. That is the same price as Claude Opus 4.8 and half of Claude Fable 5's $10/$50 rate, according to Anthropic's July 24, 2026 launch pricing. Anthropic's pricing page still lists those rates as of September 26, 2026.

Is Claude Opus 5.5 cheaper than Opus 5?

Yes. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5's $5/$25. Cache reads fall from $0.50 to $0.20 per million tokens and batch pricing from $2.50/$12.50 to $2/$10. Opus 5.5 has four breaking changes, so test before you switch model IDs.

Should I use Fable 5.1 or Opus 5.5?

Start with Opus 5.5. Anthropic's models overview recommends it for most workloads and reserves Fable 5.1 for demanding reasoning and long-horizon agentic work, or for cases where your evals on Opus 5.5 at higher effort still fall short. Fable 5.1 costs $10/$50 per million tokens, 2.5 times Opus 5.5's $4/$20.

What does a typical coding session cost on Opus 5 vs Opus 5.5?

Using Anthropic's worked example of 50,000 input and 15,000 output tokens, token cost is $0.625 on Opus 5 and $0.50 on Opus 5.5. With 40,000 of those input tokens served from cache, it drops to $0.445 and $0.348. Your real cost depends on how many tokens each model uses to finish the task.

How much do 5 million tokens cost on Claude Opus 5?

At standard API rates, 5 million input tokens cost $25 on Opus 5 and 5 million output tokens cost $125. On Opus 5.5 the same volumes cost $20 and $100. The Batch API halves every line, so 5 million input tokens on Opus 5 cost $12.50 in a batch job. Real workloads mix the two: 4 million input plus 1 million output tokens is $45 on Opus 5 and $36 on Opus 5.5, before cache reads.

Is Claude Opus 5 cheaper than Claude Fable 5?

Yes, on a per-token basis Opus 5 is exactly half the price of Fable 5. On a per-task basis for computer-use workloads, Anthropic says Opus 5 beats Fable 5 at roughly one third of the cost on the OSWorld 2.0 benchmark, an even larger practical gap than the token price alone suggests.

Does Claude Opus 5 cost more than Claude Opus 4.8?

No. Opus 5 and Opus 4.8 are priced identically at $5 per million input tokens and $25 per million output tokens. Anthropic did not raise the price at the top non-Fable tier for this release.

What is the cheapest way to use Claude Opus 5?

The standard API rate is $5/$25 per million tokens. The Batch API halves that to $2.50/$12.50 for work that can wait, and prompt cache reads cost $0.50 per million tokens instead of $5, per Anthropic's pricing page. If you are not tied to Opus 5, Opus 5.5 at $4/$20 is cheaper on every line. Actual cost per task can still vary significantly based on how many turns or retries a workload needs, which is why Anthropic frames some of its Opus 5 claims in cost-per-task terms rather than token price alone.

Should I switch from Fable 5 to Opus 5 to save money?

For most general workloads, yes, since on the Artificial Analysis Intelligence Index v4.3.2 Opus 5 scores 51 against Fable 5's 50 (#11 of 211, checked September 26, 2026) at half the token price. If you are moving off Fable 5 today, also test Opus 5.5, which Anthropic now recommends for most workloads at $4/$20. The exception is cybersecurity-specific work, where Anthropic's own disclosure places Opus 5 behind Mythos 5, so that is not a workload to migrate on price alone without testing quality first.

Why did Anthropic price Opus 5 the same as Opus 4.8 instead of higher?

Anthropic has not published its reasoning for the price point. What is confirmed is that Opus 5 matches Opus 4.8's $5/$25 rate rather than introducing a new, higher tier, while Fable 5 remains priced at $10/$50.

How does Opus 5's price compare across different task types?

The $5/$25 rate is flat across task types, but effective cost per task is not. Anthropic's disclosed benchmarks suggest the price advantage compounds on computer-use and automation workloads (OSWorld 2.0, Zapier's AutomationBench), where fewer retries mean fewer total tokens billed, on top of the lower per-token rate.

---

Sources: Anthropic pricing page (checked September 26, 2026), Anthropic models overview (checked September 26, 2026), Claude Opus 5.5 model page, Anthropic, Introducing Claude Opus 5, CNBC, Anthropic's Claude Opus 5 launch, TechCrunch, Anthropic launches Opus 5, Artificial Analysis Intelligence Index