Claude Opus 5.5 price cut and the benchmarks, checked
Claude Opus 5.5 price drops to $4 and $20 per million tokens. What the cut saves, which Fable-beating scores are Anthropic's own, and where to use it.
Source-based. Written from the documents, reporting and reviews linked in the text. Nothing here was tested hands-on by The Ruling Desk. How we work
Anthropic released Claude Opus 5.5 on September 22, 2026, and the Claude Opus 5.5 price is lower than its predecessor's: $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Anthropic says the new model performs at the level of its pricier Claude Fable 5.1 on most work. That claim rests mostly on Anthropic's own tests, and one independent index so far points the same way.
Key takeaways
- List prices fall 20% on input and output, and cache reads fall 60%, from $0.50 to $0.20 per million tokens, according to Anthropic's pricing docs as of September 23, 2026.
- Anthropic's own benchmark table has Opus 5.5 ahead of Fable 5.1 on all nine results it lists for both. Those are Anthropic's runs, except AutomationBench, which Zapier ran.
- Independently, Artificial Analysis scores Opus 5.5 at 58 on its Intelligence Index at max effort, above Fable 5.1's 53, and measures it at 76 to 92 tokens per second, against 53 to 57 for Opus 5.
- It's on the Claude API as
claude-opus-5-5and on Amazon Bedrock, Google Cloud and Microsoft Foundry. - Sonnet 5.5 and Haiku 5.5 are due "in the coming weeks," with no date or price yet.
What the Claude Opus 5.5 price cut saves in dollars
Every line on Anthropic's pricing page moved down. All prices are in US dollars per million tokens, as listed on September 23, 2026:
| Item | Opus 5 | Opus 5.5 | Fable 5.1 |
|---|---|---|---|
| Input | $5 | $4 | $10 |
| Output | $25 | $20 | $50 |
| Cache write (5 minutes) | $6.25 | $5 | $12.50 |
| Cache write (1 hour) | $10 | $8 | $20 |
| Cache read | $0.50 | $0.20 | $0.25 |
| Batch input / output | $2.50 / $12.50 | $2 / $10 | $5 / $25 |
| Fast mode input / output | $10 / $50 | $8 / $40 | Not listed |
To put that in plain numbers, here's our arithmetic at list prices: a job that sends 10 million input tokens and gets back 2 million output tokens costs $100 on Opus 5, $80 on Opus 5.5 and $200 on Fable 5.1. The cache read cut is the biggest in percentage terms, which matters most for coding agents that resend the same large context over and over.
Anthropic claims a bigger saving than the price table shows. Its announcement says Opus 5.5 "costs less per token than Opus 5 and uses fewer tokens per task," which it says "nets out to a 40% drop in costs" on typical workloads at default settings. Only the 20% is on the price list. The rest depends on Opus 5.5 using fewer tokens for your prompts, which you can only confirm on your own bills.
Claude Opus 5.5 benchmarks: whose numbers are whose
The headline "beats Fable" results come from Anthropic's table. Its footnotes say Opus 5.5 ran at max effort unless noted (Terminal-Bench 4.0 at xhigh), GPT figures are "as reported by OpenAI," and AutomationBench "was run and reported by Zapier."
| Benchmark (Anthropic's table) | Opus 5.5 | Fable 5.1 | Opus 5 |
|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 66.4% | 55.8% | 52.3% |
| FrontierCode v1.1 | 54.4% | 50.3% | 48.0% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% |
| GDPval-AA v2.1 (knowledge work, Elo) | 1846 | 1735 | 1708 |
| AutomationBench (run by Zapier) | 40.0% | 31.4% | 26.9% |
| Humanity's Last Exam, with tools | 67.7% | 65.6% | 63.6% |
| Terminal-Bench-Science 0.1 | 58.7% | 52.6% | 29.0% |
| OSWorld 2.0 (computer use) | 81.8% | 80.7% | 74.0% |
| Chartography, with tools | 89.0% | 88.4% | 83.4% |
Some of those gaps are wide, like 10.6 points on Terminal-Bench. Others, like OSWorld and Chartography, are about a point, which is close to a tie. The same table also shows OpenAI's GPT-6 Astra ahead of Opus 5.5 on AutomationBench (41.4%) and Terminal-Bench-Science (64.6%), using OpenAI's reported figures.
What independent testing shows so far
Artificial Analysis, which runs its own evaluations, scores Opus 5.5 at 58 on its Intelligence Index v4.3.2 at max effort and 54 at high effort. On the same chart, as of September 23, 2026, Fable 5.1 tops out at 53, GPT-6 Astra at 53 and Opus 5 at 51. That supports Anthropic's broad claim, but it's a single composite index from one outfit, not a check of each number above.
The speed claim holds up too. Anthropic says Opus 5.5 generates output "more than 30% faster than Opus 5"; Artificial Analysis measured 76 to 92 output tokens per second across effort levels, against 53 to 57 for Opus 5. It also lists a cost of $0.55 to $5.98 per index task depending on effort, which is a reminder that higher effort settings burn far more tokens.
Where you can use Claude Opus 5.5
According to the model overview, Opus 5.5 is on the Claude API, Amazon Bedrock (anthropic.claude-opus-5-5), Google Cloud, Microsoft Foundry and Claude Platform on AWS, with a 1 million token context window and up to 128,000 output tokens. Anthropic commits to keeping it available until at least September 22, 2027.
In the Claude apps, Anthropic says it raised five-hour usage limits on the Pro, Max, Team and seat-based Enterprise plans. Fast mode, a research preview, runs Opus 5.5 at up to 2.5 times the speed in Claude Code and on the API, at $8 and $40 per million tokens.
If you're moving code over from Opus 5, read the migration notes first. The docs list four breaking changes, including that thinking can't be turned off and that forcing a specific tool call now returns an error.
What the Sonnet 5.5 and Haiku 5.5 timing means
Anthropic says Sonnet 5.5 and Haiku 5.5 "will follow in the coming weeks, with many of the same improvements to performance, efficiency, and safety." There's no date and no price. Today the cheaper tiers are Sonnet 5 at $2 and $10 and Haiku 4.5 at $1 and $5.
If you run a high-volume app on Sonnet or Haiku, the practical move is to wait a few weeks rather than jump to Opus 5.5. If you already pay for Opus 5 or Fable 5.1, you can test Opus 5.5 now. The competition moved the same day: OpenAI released GPT-6 Sol and Luna with price cuts of about half, so it's worth pricing both before you commit.
TechCrunch notes this is Anthropic's first release since CEO Dario Amodei committed to pacing capability gains so alignment research can keep up. Anthropic says METR and Frontier Design evaluated the model before release.
Bottom line
Opus 5.5 is 20% cheaper per token than Opus 5 on the price list, and cache reads are 60% cheaper. The claim that it beats Fable 5.1 comes from Anthropic's own benchmark table; Artificial Analysis' independent index and speed tests agree with the direction, but we found no outside reproduction of the individual scores as of September 23, 2026. If you use Opus 5, test the switch on a real workload and watch your token counts. If you use Sonnet or Haiku, wait for the 5.5 versions. More coverage lives in our AI section.
FAQ
How much does Claude Opus 5.5 cost?
The Claude Opus 5.5 price on the Claude API is $4 per million input tokens and $20 per million output tokens, as of September 23, 2026. Cache reads cost $0.20 per million, and the Batch API halves input and output prices. Bedrock and Google Cloud publish their own pricing.
Is Claude Opus 5.5 better than Fable 5.1?
In Anthropic's own benchmark table it scores higher on all nine results listed for both, some by a point or less. Artificial Analysis' independent Intelligence Index also ranks it above Fable 5.1, 58 to 53 at max effort. Fable 5.1 still costs 2.5 times as much per token.
When will Claude Sonnet 5.5 and Haiku 5.5 come out?
Anthropic only says "in the coming weeks." It hasn't announced a date or prices for either model.