Claude Opus 5.5 Review: Fable-Level Results for Less

Claude Opus 5.5 is out. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads.

The per-token price cut is 20%, so the 40% figure has to come from more than the rate card. We broke down where the rest comes from and priced an illustrative agentic coding session on both models. The short version is that the 40% holds for cache-heavy agentic work, and almost all of it comes from the 60% cut to cache reads. A short chat with little cached context saves closer to 20%.

Price changes

Input, output, and cache writes all fall by the same 20%. Cache reads fall by 60%. Context window and max output are unchanged at 1,000,000 and 128,000 tokens. We cover the full rate card for every current model in our Claude API pricing guide.

The cut also changes the ratio between cache reads and regular input. On Opus 5, a cache read cost 10% of the input price. On Opus 5.5, it costs 5%. Cache writes stay at 1.25× the input price on both models.

Three sources of savings

Anthropic's announcement points to three separate changes. First, every token type is at least 20% cheaper. Second, cache reads are 60% cheaper, which applies mostly to agentic work, where files, tool definitions, and conversation history are resent as cached context on every turn. Third, Anthropic says Opus 5.5 uses fewer tokens per task than Opus 5, though it hasn't published a typical figure for that reduction.

The first two follow directly from the price table. The third depends on the task.

Example session

To see how these combine, we priced one illustrative agentic coding session on both models.

Assumptions:

• 80 turns
• 60,000 cache-read tokens per turn (repo files, tools, and history)
• 1,000 fresh input tokens per turn
• 2,000 cache-write tokens per turn (new context added to the cache)
• Opus 5: 1,000 output tokens per turn
• Opus 5.5: 750 output tokens per turn, 25% fewer. This is our assumption for illustration, not a figure from Anthropic.

Across the session, that comes to 4.8M cache-read tokens, 80,000 input tokens, and 160,000 cache-write tokens on both models. Output is 80,000 tokens on Opus 5 and 60,000 on Opus 5.5.

Opus 5:

• Cache reads: 4.8M × $0.50/M = $2.40
• Input: 0.08M × $5.00/M = $0.40
• Cache writes: 0.16M × $6.25/M = $1.00
• Output: 0.08M × $25.00/M = $2.00
• Total: $5.80

Opus 5.5, same token counts (price changes only):

• Cache reads: 4.8M × $0.20/M = $0.96
• Input: 0.08M × $4.00/M = $0.32
• Cache writes: 0.16M × $5.00/M = $0.80
• Output: 0.08M × $20.00/M = $1.60
• Total: $3.68 (37% less than Opus 5)

Opus 5.5, with 25% fewer output tokens:

• Output: 0.06M × $20.00/M = $1.20
• Total: $3.28 (43% less than Opus 5)

In this session, the new prices alone account for a 37% reduction, most of it from cache reads. Cache reads were the largest line item on Opus 5 at $2.40 of $5.80, and they took the largest cut. The assumed drop in output tokens adds another 6 points.

A few factors would change these results. A short exchange with little cached context would land closer to 20%, since the cache-read discount barely applies. A task where Opus 5.5 also needs fewer turns would save more, because every turn removed also removes its cache reads. We kept the turn count the same on both models to isolate the pricing effect.

An early tester's result

Anthropic cited one early tester who audited and fixed a 200,000-line codebase in under three hours with Opus 5.5, versus more than 20 hours with Opus 5, while using 2.5 times fewer tokens.

Using 40% of the tokens at 80% of the per-token price gives 0.4 × 0.8 = 0.32, or about 68% lower cost for that task. This is a single reported case, not a typical result.

Compared with Fable 5.1

Since Anthropic positions Opus 5.5 at Fable 5.1's level on most work, we also compared the two on price. We covered Fable 5.1's own pricing changes in our Claude Fable 5.1 review.

The gap is large on input and output and small on cache reads. Applying the example session's Opus 5 token counts to both models, and leaving out cache writes, Fable 5.1 comes to $6.00 (cache reads $1.20, input $0.80, output $4.00) and Opus 5.5 comes to $2.88. That is 52% less. The more of a workload sits in cache reads, the closer the two models get in cost.

Fast mode

Opus 5.5 has a fast mode in Claude Code and the Claude Platform, priced at $8 per million input tokens and $40 per million output tokens. That is twice the standard rate, for up to 2.5× faster output. The cost per task doubles for the input and output portion, and the time saved depends on how close a given workload gets to the 2.5× figure.

Checking it yourself

Every response includes token usage by type, so you can compare the two models on your own tasks and replace the 25% assumption above with your own figure. Puter.js gives free access to Opus 5.5 without an API key, and the same call works for claude-opus-5.

Script tag:

<html>
<body>
  <script src="https://js.puter.com/v2/"></script>
  <script>
    puter.ai.chat("Write a function that debounces API calls with cancellation support", {
      model: "claude-opus-5-5"
    }).then(response => {
      puter.print(response.message.content[0].text);
      console.log(response.usage);
    });
  </script>
</body>
</html>

npm (@heyputer/puter.js):

import { puter } from '@heyputer/puter.js';

const response = await puter.ai.chat(
  "Write a function that debounces API calls with cancellation support",
  { model: 'claude-opus-5-5' }
);

console.log(response.message.content[0].text);
console.log(response.usage);

Run the same task several times per model and compare the average output tokens. If you'd rather use the official SDK, Puter also exposes an Anthropic-compatible endpoint that returns the same usage fields.

What this adds up to

On our example session, Anthropic's 40% figure checks out. The new prices alone bring the session from $5.80 to $3.68, a 37% reduction, before assuming any change in how many tokens the model uses. Adding our assumed 25% drop in output tokens takes it to 43%.

Nearly all of that comes from one line item. Cache reads were 41% of the Opus 5 bill and took a 60% cut, while everything else fell 20%. The "fewer tokens per task" claim, which we did not measure, moves the result by a few points either way.

That also sets the limit of the claim. The 40% applies to workloads where most of the spend is cached context, which is what agentic coding sessions look like. A short exchange with little repeated context gets the 20% rate cut and not much else. The same shape shows up against Fable 5.1, where Opus 5.5 is 52% cheaper on our session but the gap narrows as the cache-read share grows, because those two prices are only $0.05 apart.

FAQ

When was Claude Opus 5.5 released? Claude Opus 5.5 was released on September 22, 2026, as the first model in Anthropic's Claude 5.5 family.

What does Claude Opus 5.5 cost? Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Cache reads are $0.20 per million tokens and cache writes are $5 per million tokens.

Is Claude Opus 5.5 cheaper than Opus 5? Claude Opus 5.5 is 20% cheaper than Opus 5 on input and output tokens and 60% cheaper on cache reads. Anthropic says typical workloads cost 40% less overall once reduced token usage is included.

Are other Claude 5.5 models coming? Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks, according to Anthropic.

Related

• Claude API pricing: full breakdown of costs
• Claude Fable 5.1 review
• Free, unlimited Claude API with Puter.js
• Puter.js AI documentation