Claude Opus 4.6 vs 4.7 vs 4.8 pricing compared

If you still run Claude Opus 4.6 on the API, the price list gives you no reason to hold back: Opus 4.7 and 4.8 charge exactly the same per-token rates. What changes is how many tokens you get billed for, how much the model thinks by default, and whether fast mode works at all. This comparison walks through each of those using only Anthropic's own pricing docs, effort docs, and launch posts.

All three Claude Opus models list at $5 input and $25 output

Anthropic has kept the standard rate flat across these three Opus releases, so each one bills $5 for every million tokens you send and $25 for every million it returns. Opus 4.7 launched at the Opus 4.6 price, and Opus 4.8 launched at the Opus 4.7 price, so the rows for these three models on the pricing page read the same.

So there is no per-model discount to chase among these three. If you are comparing against the newer Opus 5.5 as well, its rates are covered in Claude Opus 5.5 pricing, benchmarks, API changes.

Batch API halves the rate to $2.50 and $12.50 on each

Batch requests cost $2.50 per million input tokens and $12.50 per million output tokens on Opus 4.8, 4.7 and 4.6 alike. Because the discount is identical, a batch pipeline has no price-list reason to stay on 4.6 either.

The tokenizer difference still reaches batch jobs, though. The discount applies per token, so a larger token count raises a batch bill the same way it raises a standard one.

The 1M context window is billed at the standard rate

All three models fall under what Anthropic calls Claude 4.6 and later, which get the full 1M-token context window at standard pricing. There is no long-context surcharge: a 900k-token request pays the same per-token rate as a 9k-token one.

For long-document workloads, then, the rate card does not separate the three models. Only the token count does, and that is where 4.6 and its successors part ways.

Claude Opus 4.7 and 4.8 bill up to 35% more tokens than 4.6

Identical rates do not guarantee identical invoices. Starting with Opus 4.7, Anthropic switched to a new tokenizer, and that change can raise your real spend when you leave Claude Opus 4.6.

The newer tokenizer maps the same text to more tokens

The pricing docs say Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. Opus 4.6, like Sonnet 4.6, stays on the previous tokenizer. The Opus 4.7 launch post gives the spread more precisely: the same input maps to roughly 1.0–1.35× as many tokens, depending on the content type.

In practice, the same prompt sent to Opus 4.6 and to Opus 4.7 can come back with a larger token count on 4.7, billed at the same $5 rate. That makes a flat 35% uplift the wrong planning number. The 1.35× figure is the top of the range Anthropic quotes, and some content lands close to 1.0×.

Moving from Opus 4.7 to 4.8 adds no new tokenizer cost

The tokenizer change happened once, at Opus 4.7. Anthropic's wording covers Claude 4.7 and later models, so Opus 4.8 shares the 4.7 tokenizer and the same text is counted the same way on both.

If you already moved from 4.6 to 4.7 and absorbed that jump, going to 4.8 does not bring a second one from tokenization. What can still shift is how many tokens the model spends working, which comes down to effort.

Default effort sets the real token cost of Claude Opus 4.8

The tokenizer decides how text is counted, while effort decides how much text the model generates on the way to an answer. On the API, effort is a per-request setting, and a request that doesn't set it gets the default.

Opus 4.8 defaults to high effort at 4.7-level token use on coding

Opus 4.8 runs at high effort unless you ask for something else. Anthropic says that on coding tasks this setting spends about as many tokens as Opus 4.7's default while performing better. Coding workloads that move from 4.7 to 4.8 on default settings shouldn't see output spend climb for that reason alone.

High is the default on every model that supports effort except Opus 5.5, and the max level is available on Opus 4.8, 4.7 and 4.6. The 4.7 launch post also notes that Opus 4.7 thinks more at higher effort levels, particularly on later turns in agentic settings. If your 4.6 workloads are long agent loops, check output tokens per run after the switch and set effort explicitly instead of inheriting the default.

Lower effort on claude.ai stretches Claude usage limits

On claude.ai the same control shows up in your usage limits rather than on an invoice. Anthropic says that at lower effort settings Claude responds faster and uses up rate limits more slowly, and the effort control is available on all plans.

If you keep running into Claude usage limits on Opus 4.8, lower the effort for routine questions; that is the setting Anthropic itself ties to slower rate-limit use. Anthropic has also raised rate limits in Claude Code to make room for the heavier token use of higher effort levels, which shows how much the effort level weighs on a plan's allowance.

Fast mode costs $10 and $50 and runs only on Claude Opus 4.8

Fast mode is the one place where the three models are priced differently. On Opus 4.8, fast mode lets the model work at 2.5× the speed for $10 per million input tokens and $50 per million output tokens, and Anthropic says that is three times cheaper than fast mode was on previous models.

The two older models handle a fast mode request in different ways. Opus 4.7 rejects speed: "fast" with an error, while Opus 4.6 accepts it but runs at standard speed and bills standard rates.

The 4.6 behavior is the one to watch. Code written for fast mode on 4.8 won't fail if you point it at 4.6. It will just run at standard speed and bill at the normal price.

Fast mode is also limited to the first-party Claude API. It is not offered on Claude Platform on AWS or on partner-operated cloud platforms, so teams buying Claude through a cloud provider should budget at standard rates. For everything else that changed in the newest model, see Claude Opus 4.8: Pricing, Fast Mode, Workflows.

Summary: same Opus prices, but token counts set the bill

Claude Opus 4.6, 4.7 and 4.8 share one price list: $5 input and $25 output, $2.50 and $12.50 on batch, and no surcharge for the 1M context window. The bill can still differ. Opus 4.7 and 4.8 count the same text as roughly 1.0–1.35× the tokens, the effort default decides how much the model generates, and fast mode exists only on 4.8.

Before you move a workload off Opus 4.6, send a sample of your own prompts to the newer model and compare the token counts. Anthropic recommends measuring the difference on real traffic, and that is the only way to learn where your content falls in that range.