API pricing comparison
The cheapest LLM API prices per token
As of 2026-06-22, the cheapest LLM API input tokens are Amazon nova-micro at $0.035 per 1M, and the cheapest output is Mistral ministral-3-3b at $0.10 per 1M. Every price is dated and first-party-sourced.
Cheapest input tokens
| # | Model | Provider | Input /1M | Source |
|---|---|---|---|---|
| 1 | nova-micro | Amazon | $0.035 | aws.amazon.com |
| 2 | command-r7b | Cohere | $0.037 | cohere.com |
| 3 | qwen-flash | Alibaba | $0.050 | www.alibabacloud.com |
| 4 | qwen-turbo | Alibaba | $0.050 | www.alibabacloud.com |
| 5 | nova-lite | Amazon | $0.060 | aws.amazon.com |
| 6 | gemini-1.5-flash | $0.075 | web.archive.org | |
| 7 | qwen3.5-flash | Alibaba | $0.10 | www.alibabacloud.com |
| 8 | gemini-2.5-flash-lite | $0.10 | web.archive.org | |
| 9 | devstral-small-2 | Mistral | $0.10 | mistral.ai |
| 10 | ministral-3-3b | Mistral | $0.10 | mistral.ai |
Cheapest output tokens
| # | Model | Provider | Output /1M | Source |
|---|---|---|---|---|
| 1 | ministral-3-3b | Mistral | $0.10 | mistral.ai |
| 2 | nova-micro | Amazon | $0.14 | aws.amazon.com |
| 3 | command-r7b | Cohere | $0.15 | cohere.com |
| 4 | ministral-3-8b | Mistral | $0.15 | mistral.ai |
| 5 | llama-3.1-8b | Meta | $0.18 | www.together.ai |
| 6 | qwen-turbo | Alibaba | $0.20 | www.alibabacloud.com |
| 7 | ministral-3-14b | Mistral | $0.20 | mistral.ai |
| 8 | nova-lite | Amazon | $0.24 | aws.amazon.com |
| 9 | deepseek-v4-flash | DeepSeek | $0.28 | api-docs.deepseek.com |
| 10 | gemini-1.5-flash | $0.30 | web.archive.org |
Cheap per token is not the same as cheap per job: the lowest-priced models are usually small, so they can spend more tokens to do the same work. And list price is not your bill once caching and short prompts are counted. To see what you actually spend, priced at the rate in effect each day, use Goei.
More from the index
- Claude vs GPT API pricing
- All per-model pricing pages
- The interactive chart
- The dataset, JSON endpoints, and how to cite it
Open data, CC BY 4.0. Prices validated as of 2026-06-22.