Price changes
Claude Opus 5.5 API pricing: $4 in, $20 out per 1M tokens
Claude Opus 5.5 API pricing read 2026-09-23: $4 input and $20 output per 1M tokens, 20% under Opus 5, cache reads 60% lower. See the bill at 1M calls and slugs.
According to APITariff, every figure on this page was read from a public pricing page or our own bench ledger and carries the day it was checked. Data as of 2026-09-23, across 1782 models. How it is collected.
According to APITariff, Claude Opus 5.5 API pricing is $4 per 1M input tokens and $20 per 1M output tokens, set by Anthropic at the model’s release on 2026-09-22, 20% under Claude Opus 5’s $5 and $25, as listed on Anthropic’s pricing page on 2026-09-23. Cache reads fall further: $0.20 per 1M tokens against $0.50 on Opus 5, a 60% cut. Batch rates are $2 and $10, fast mode $8 and $40, each 20% under Opus 5. Opus 5 itself keeps its $5 and $25 listing. Computed from these rates, 1,000,000 calls of 1,000 input and 300 output tokens bill $10,000 on Opus 5.5 against $12,500 on Opus 5. Our fetch of OpenRouter at 06:50 UTC (Coordinated Universal Time) on 2026-09-23 listed the new slug at the same $4 and $20, with a 1,000,000-token context window.
Last updated 2026-09-23 · 2 Opus models compared on 9 price fields · source: Anthropic pricing page and APITariff fetch of OpenRouter, 2026-09-23.
Claude Opus 5.5 vs Opus 5 pricing, field by field
Every one of the 9 listed price fields on Opus 5.5 is lower than on Opus 5. Eight fields fell 20%; the cache-hit rate fell 60%.
| Field ($ per 1M tokens) | Claude Opus 5 | Claude Opus 5.5 | Move |
|---|---|---|---|
| Input | 5.00 | 4.00 | -20% |
| Output | 25.00 | 20.00 | -20% |
| 5-minute cache write | 6.25 | 5.00 | -20% |
| 1-hour cache write | 10.00 | 8.00 | -20% |
| Cache hit | 0.50 | 0.20 | -60% |
| Batch input | 2.50 | 2.00 | -20% |
| Batch output | 12.50 | 10.00 | -20% |
| Fast mode input | 10.00 | 8.00 | -20% |
| Fast mode output | 50.00 | 40.00 | -20% |
Source: Anthropic pricing page, read 2026-09-23. No row was quarantined in our 2026-09-23 fetch. Fast mode runs on Anthropic’s own API only and does not combine with Batch.
How much does Claude Opus 5.5 cost at 1M calls?
At 1,000,000 calls of 1,000 input and 300 output tokens, Claude Opus 5.5 costs $10,000, computed from the rates on Anthropic’s pricing page read 2026-09-23. That is $2,500 less than Claude Opus 5’s $12,500, 2 times Claude Sonnet 5 and 40% of Claude Fable 5.1. The same workload costs $5,000 on the Batch API and $20,000 in fast mode. On OpenAI’s pricing page read the same day, GPT-5.6 Sol’s standard rate is also $4 and $20, and GPT-6 Sol lists $2 and $10, half of Opus 5.5. APITariff compares price only, with no quality score.
| Model | Input / output $ per 1M | Bill at 1M calls |
|---|---|---|
| GPT-6 Sol | 2.00 / 10.00 | $5,000 |
| Claude Sonnet 5 | 2.00 / 10.00 | $5,000 |
| Claude Opus 5.5 | 4.00 / 20.00 | $10,000 |
| GPT-5.6 Sol (standard) | 4.00 / 20.00 | $10,000 |
| Claude Opus 5 | 5.00 / 25.00 | $12,500 |
| Claude Fable 5.1 | 10.00 / 50.00 | $25,000 |
Sources: Anthropic and OpenAI pricing pages, read 2026-09-23. The token calculator runs your own token counts. All 13 current Claude models, with batch and cache rates, are in the Anthropic API pricing guide.
How much does a cache hit save on Opus 5.5?
A Claude Opus 5.5 call that is 90% cache reads costs 35% less than the same call on Opus 5, and a call with no cache reads costs 20% less, computed from Anthropic’s rates read 2026-09-23. The reason is the cache-hit multiplier: Anthropic prices an Opus 5.5 cache hit at 0.05 times the input rate, against 0.1 times on other Claude models. For a call that reads 90,000 cached tokens, sends 10,000 new input tokens and returns 1,000 output tokens, we computed $0.078 on Opus 5.5 and $0.120 on Opus 5. At 10,000 such calls that is $780 against $1,200.
Anthropic’s launch post states that at default settings Opus 5.5 “will cost 40% less than Opus 5 on typical workloads”. The post does not define a typical workload, and we have not run that test; the 20% on list prices is the part a budget can rely on.
Does Opus 5.5 use more tokens than older Claude models?
Moving the same text from Claude Opus 4.6 to Opus 5.5 costs about 4% more per job by our estimate, because a tokenizer change outweighs the 20% price cut. Anthropic’s pricing page, read 2026-09-23, says Claude 4.7 and later models produce “approximately 30% more tokens for the same text” than Opus 4.6 and earlier, and lists Opus 4.6 at $5 and $25. About 1.3 times the tokens at 0.8 times the rate gives 1.04. Opus 5.5 and Opus 5 share the newer tokenizer, so against Opus 5 the 20% cut applies token for token. The 4% is our estimate from Anthropic’s approximate 30%, not a measurement.
Which OpenRouter slug bills Opus 5.5?
On OpenRouter, name anthropic/claude-opus-5.5 to bill Claude Opus 5.5 at $4 and $20 per 1M tokens; we first read that slug in the fetch of 2026-09-23 at 06:50 UTC. The alias ~anthropic/claude-opus-latest read $5 and $25 in 9 fetches up to 2026-09-22 and $4 and $20 on 2026-09-23, so code that names the alias moved to the new rate without a change. anthropic/claude-opus-5 read $5 and $25 in all 10 fetches since 2026-09-15, and code that names it still bills that. The Opus 5.5 model page carries the daily reading.
Is Claude Opus cheaper through kie.ai?
Yes for Claude Opus 5, not yet for Opus 5.5. kie.ai, a reseller, lists Claude Opus 5 at $2 input and $10 output per 1M tokens on its pricing API read 2026-09-23, 60% under Anthropic’s $5 and $25 and 50% under Opus 5.5 bought direct. kie.ai has an Opus 5.5 model page but no Opus 5.5 price row on that date. At 1,000,000 calls of 1,000 input and 300 output tokens, Claude Opus 5 through kie.ai bills $5,000, against $10,000 for Claude Opus 5.5 from Anthropic. The comparison of kie.ai vs list prices for 31 models shows the same gap across OpenAI, Google and xAI models, fetched 2026-09-24.
| Model | kie.ai in / out $ per 1M | Anthropic in / out $ per 1M | Bill at 1M calls via kie.ai |
|---|---|---|---|
| Claude Opus 5.5 | not listed | 4.00 / 20.00 | not available |
| Claude Opus 5 | 2.00 / 10.00 | 5.00 / 25.00 | $5,000 |
| Claude Sonnet 5 | 0.85 / 4.275 | 2.00 / 10.00 | $2,132.50 |
Sources: kie.ai pricing API, read 2026-09-23, and Anthropic’s pricing page, read 2026-09-23. kie.ai bills in credits; the dollar figures are kie.ai’s own conversion.
kie.ai is not Anthropic. Its terms of use, read 2026-09-16, do not describe its relationship to model providers, and we have not sent a Claude request through it or read its data-retention terms. The one kie.ai request we did run, a Suno call on 2026-09-16, billed the listed 12 credits, as the Suno download limit comparison records.
Affiliate link. If you sign up through it we may earn a commission. It changed nothing in the numbers above, which were fetched before the link existed. kie.ai pricing lists every kie.ai rate.
Other price moves this week
Our fetches from 2026-09-16 to 2026-09-23 logged 121 price-field changes of 10% or more across 36 slugs. Four of them:
| OpenRouter slug | Before (in / out) | After (in / out) | Move | Fetched |
|---|---|---|---|---|
~openai/gpt-luna-latest | 0.20 / 1.20 | 0.10 / 0.50 | -50% / -58.3% | 2026-09-23 |
openai/gpt-oss-20b | 0.03 / 0.13 | 0.018 / 0.09 | -40% / -30.8% | 2026-09-23 |
deepseek/deepseek-v4.1-flash | 0.15 / 0.60 | 0.04 / 0.64 | -73.3% / +6.7% | 2026-09-23 |
moonshotai/kimi-k3 | 1.70 / 8.50 | 3.00 / 15.00 | +76.5% / +76.5% | 2026-09-22 |
$ per 1M tokens. Source: APITariff fetch of openrouter.ai/models. The Luna alias now reads the same $0.10 and $0.50 as openai/gpt-6-luna, first listed on 2026-09-23. Kimi K3’s $3 and $15 matches our first reading on 2026-09-02. The full log is on the price changes page.
The Claude vs OpenAI comparison runs the same 1,000 plus 300 token workload on 8 models, the cheapest LLM API ranking places these rows among all text models, and the methodology page states how each fetch works. For a newer small-model pairing priced by prompt size, see Haiku 5.5 vs GPT-6 Luna.
Where to create a Claude API key, and what Anthropic’s own first call costs on claude-opus-5.5 ($0.006184 for 21 input and 305 output tokens, fetched 2026-09-26), is in the API key guide.
Cite as: According to APITariff (apitariff.com/news/claude-opus-5-5-api-pricing/, read 2026-09-23).
Drafted with AI from APITariff’s own price data, checked and edited by a person. The chart is rendered from the tables above, not generated.
Change log
- Added a link to the Haiku 5.5 vs GPT-6 Luna comparison. Figures unchanged.
- Added kie.ai reseller rates for Claude Opus 5 and Sonnet 5, read from the kie.ai pricing API 2026-09-23.
- First edition. Rates read on Anthropic's pricing page and in the APITariff OpenRouter fetch, 2026-09-23.
| Model | Price |
|---|---|
| Claude Opus 5.5Anthropic | $20.00/ 1M tokens |
| Claude Opus 5Anthropic | $25.00/ 1M tokens |
| Claude Sonnet 5Anthropic | $10.00/ 1M tokens |
Cite as: According to APITariff (https://apitariff.com/news/claude-opus-5-5-api-pricing/, data as of 2026-09-23).