Guides
How much do 1,000 tokens cost? 12 models
How much do 1,000 tokens cost? From $0.00003 to $0.60 per 1,000 output tokens across 12 text models, fetched from OpenRouter on 2026-10-14, with the working.
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 2026-10-14, across 1795 models. How it is collected.
According to APITariff, 1,000 tokens cost between $0.00003 and $0.60 on output and between $0.000018 and $0.15 on input across the 12 text models compared here, from OpenRouter rates and the Anthropic pricing page fetched on 2026-10-14. 1,000 input tokens cost $0.0001 on Claude Haiku 5.5, $0.002 on Claude Sonnet 5.5 and $0.15 on OpenAI o1-pro. 1,000 output tokens cost $0.0005, $0.01 and $0.60 on the same three. Because providers quote a price per 1M tokens, you divide by 1,000 to get your number: $2.00 per 1M is $0.002 per 1,000. Among the 12 models shown, the widest gap is on output. There o1-pro costs 20,000 times what Mistral Nemo costs: $0.60 against $0.00003. Each provider counts tokens its own way. So 1,000 tokens are not the same amount of text on every row.
Last updated 2026-10-14 · 12 models · 8 providers · sources: APITariff daily fetch of OpenRouter, Anthropic API pricing page, kie.ai pricing rows.
How much is 1,000 tokens in dollars on each of the 12 models?
The 4 current Claude models cost $0.0001 to $0.01 per 1,000 input tokens and $0.0005 to $0.05 per 1,000 output tokens on 2026-10-14.
| Model | Provider | Context | Input per 1,000 | Output per 1,000 | Input / output per 1M |
|---|---|---|---|---|---|
| Claude Haiku 5.5 | Anthropic | 1,000,000 | $0.0001 | $0.0005 | 0.10 / 0.50 |
| Claude Sonnet 5.5 | Anthropic | 1,000,000 | $0.002 | $0.01 | 2.00 / 10.00 |
| Claude Opus 5.5 | Anthropic | 1,000,000 | $0.004 | $0.02 | 4.00 / 20.00 |
| Claude Fable 5.1 | Anthropic | 1,000,000 | $0.01 | $0.05 | 10.00 / 50.00 |
| gpt-oss-20b | OpenAI | 131,072 | $0.000018 | $0.00009 | 0.018 / 0.09 |
| Gemma 3 4B | 131,072 | $0.00005 | $0.0001 | 0.05 / 0.10 | |
| DeepSeek V3.2 Exp | DeepSeek | 163,840 | $0.00027 | $0.00041 | 0.27 / 0.41 |
| Qwen3.7 Flash | Alibaba | 1,000,000 | $0.00003 | $0.00013 | 0.03 / 0.13 |
| Grok Build 0.1 | xAI (listed as SpaceXAI) | 256,000 | $0.001 | $0.002 | 1.00 / 2.00 |
| Llama 3.1 8B Instruct | Meta | 131,072 | $0.00005 | $0.00008 | 0.05 / 0.08 |
| Mistral Nemo | Mistral | 131,072 | $0.000029 | $0.00003 | 0.029 / 0.03 |
| OpenAI o1-pro | OpenAI | 200,000 | $0.15 | $0.60 | 150.00 / 600.00 |
12 of 279 qualifying models shown. Source: APITariff daily fetch of OpenRouter, 2026-10-14 07:10 UTC; the 4 Claude rows match the Anthropic API pricing page read the same day. Prices are standard, with no caching or batch.
How did we pick these 12 models?
The rule is: the 4 current Claude models, the lowest output price with at least 128k context from each of 7 other providers, and the single highest output price in the set.
We counted 279 OpenRouter text models with 128k or more context, a price above $0, and no : variant suffix in the slug. The 7 other providers are OpenAI, Google, DeepSeek, Qwen, xAI, Meta and Mistral; image models and $0 rows are excluded, so a free tier never ranks. o1-pro is the highest output price among the 279. The 12 are a spread of price points, not a list of recommendations, and we run no quality bench on any of them.
How do you turn a per-1M price into a cost per call?
A call of 1,000 input and 300 output tokens costs $0.005 on Claude Sonnet 5.5, $0.00025 on Claude Haiku 5.5 and $0.33 on o1-pro.
When rates are in dollars per 1M tokens, cost per call = (1,000 x input rate + 300 x output rate) / 1,000,000.
| Model | Cost per call | Calls for $100 |
|---|---|---|
| Claude Haiku 5.5 | $0.00025 | 400,000 |
| Claude Sonnet 5.5 | $0.005 | 20,000 |
| Claude Opus 5.5 | $0.01 | 10,000 |
| Claude Fable 5.1 | $0.025 | 4,000 |
| Mistral Nemo | $0.000038 | 2,631,579 |
| OpenAI o1-pro | $0.33 | 304 |
Computed by APITariff on 2026-10-14 from the rates above. The token calculator runs the same sum for your own token counts.
Is a token the same size on every model?
No. Anthropic’s pricing page, read on 2026-10-14, says Claude 4.7 and later models split the same text into “approximately 30% more tokens” than earlier ones. By that figure, 1,000 tokens on a newer Claude model cover about 77% of the text that 1,000 tokens covered on an earlier Claude model (1 / 1.3). We derived the 77%; we did not measure it. Count your own text with the token counter: it counts OpenAI models exactly and marks Claude, Gemini and DeepSeek counts as estimates.
How many tokens are in 1,000 words, and what do they cost?
A 1,000-word English prompt is about 1,330 tokens on OpenAI’s o200k tokenizer. Claude’s own count for the same words is not measured here. At 1,330 tokens, input would cost $0.00266 on Claude Sonnet 5.5 and $0.000133 on Claude Haiku 5.5.
The 1,330 comes from the APITariff token counter. It states about 1,330 tokens per 1,000 English words on OpenAI’s o200k tokenizer, roughly 0.75 words per token, and we did not measure it in this run. Code, non-English text, numbers and JSON use more tokens per word. The Claude dollar figures are arithmetic on that OpenAI count, 1.33 x the per-1,000 input price. Claude counts differ, as the token-size section explains.
Which of the 12 models does kie.ai list?
kie.ai lists 3 of the 12 models: Claude Haiku 5.5 at $0.00006 input and $0.0003 output per 1,000 tokens, Sonnet 5.5 at $0.0008 and $0.004, and Opus 5.5 at $0.0016 and $0.008, read on 2026-10-14.
Those three sit at 40% to 60% of the direct rates above. kie.ai lists no row for the other 9, including Claude Fable 5.1 (it lists Fable 5) and o1-pro (checked 2026-10-14). kie.ai is not always the lowest route either: the lowest output price among its 38 text rows fetched on 2026-10-14 is GPT-6 Luna at $0.15 per 1M tokens, and on OpenRouter Mistral Nemo ($0.03), Llama 3.1 8B ($0.08), gpt-oss-20b ($0.09), Gemma 3 4B ($0.10) and Qwen3.7 Flash ($0.13) all list lower output prices.
Affiliate link. If you sign up through it we may earn a commission. The rates above come from kie.ai rows in our daily fetch, not from the link. The full rate list is on kie.ai’s own site. kie.ai is a reseller, not the model makers, and we have not sent a request through it or read its data-retention terms.
What is not on this page
- A words-per-token ratio measured by us: OpenAI’s token help page returned HTTP 403 on 2026-10-14. The 1,330 tokens per 1,000 English words above is the figure our own token counter states for OpenAI’s o200k tokenizer; other tokenizers differ.
- Cached input, batch and long-context rates, except one: Claude Haiku 5.5 prompts over 100,000 tokens bill $0.50 input and $2.50 output per 1M, five times the figures in the table.
- Image, video and voice pricing, which is per image, per second or per request, not per token.
- Quality, speed or benchmark scores for any of the 12 models.
- Providers’ own prices where they differ from OpenRouter’s; the 4 Claude rows are the only ones checked against the provider page.
Claude cache and batch rates are in the Anthropic API pricing guide. Weighing a Claude plan against token spend is a separate sum, worked in Claude Max vs API cost.
Drafted with AI from APITariff’s own price data, checked and edited by a person. The tables are typed from the rates above; no image on this page is AI-generated.
Change log
- First edition. 12 text models, rates fetched 2026-10-14.
| Model | Price |
|---|---|
| Claude Haiku 5.5Anthropic | $0.500/ 1M tokens |
| Claude Sonnet 5.5Anthropic | $10.00/ 1M tokens |
| Claude Opus 5.5Anthropic | $20.00/ 1M tokens |
| Claude Fable 5.1Anthropic | $50.00/ 1M tokens |
Cite as: According to APITariff (https://apitariff.com/guides/how-much-do-1000-tokens-cost/, data as of 2026-10-14).