Cheapest LLM API 2026
The cheapest LLM API on our November 1, 2026 check was Ministral 3 3B, at $0.10 per million input tokens and $0.10 per million output tokens. Below, all 36 models we track from 5 providers are ranked by input plus output price, standard tier, in US dollars, and by what one coding agent step costs. Every price comes from the provider's own pricing page.
Cheapest LLM APIs per 1M tokens
The 10 lowest by input price plus output price per 1M tokens.
| Model | Provider | Input | Cached | Output | Context |
|---|---|---|---|---|---|
Ministral 3 3Bministral-3b-2512 | Mistral | $0.10 | $0.01 | $0.10 | 256K |
Ministral 3 8Bministral-8b-2512 | Mistral | $0.15 | $0.015 | $0.15 | 256K |
Ministral 3 14Bministral-14b-2512 | Mistral | $0.20 | $0.02 | $0.20 | 256K |
GPT-6 Lunagpt-6-luna | OpenAI | $0.10 | $0.01 | $0.50 | 1.05M |
Claude Haiku 5.5claude-haiku-5-5Over 100K tokens: $0.50 in, $2.50 out. | Anthropic Claude | $0.10 | $0.01 | $0.50 | 1M |
Mistral Small 4mistral-small-2603 | Mistral | $0.15 | $0.015 | $0.60 | 256K |
Codestralcodestral-2508 | Mistral | $0.30 | $0.03 | $0.90 | 128K |
GPT-5.6 Lunagpt-5.6-lunaOver 272K tokens: $0.40 in, $1.80 out. | OpenAI | $0.20 | $0.02 | $1.20 | Not listed |
GPT-5.4 nanogpt-5.4-nanoMax input 272K tokens. | OpenAI | $0.20 | $0.02 | $1.25 | 400K |
DeepSeek V4.1 Flashdeepseek-flashPeak price. Off-peak: $0.15 in, $0.60 out. Max output 384K tokens. | DeepSeek | $0.30 | $0.006 | $1.20 | 1M |
Cheapest LLM API for a coding agent step
One step of a coding agent often sends a large prompt and gets a short answer. Here a step is 50,000 input and 2,000 output tokens, with no cache. This ranks price only, not code quality.
| Model | Provider | Per step | Per 1,000 steps |
|---|---|---|---|
| Ministral 3 3B | Mistral | $0.0052 | $5.20 |
| GPT-6 Luna | OpenAI | $0.0060 | $6.00 |
| Claude Haiku 5.5 | Anthropic Claude | $0.0060 | $6.00 |
| Ministral 3 8B | Mistral | $0.0078 | $7.80 |
| Mistral Small 4 | Mistral | $0.0087 | $8.70 |
| Ministral 3 14B | Mistral | $0.010 | $10.40 |
| GPT-5.6 Luna | OpenAI | $0.012 | $12.40 |
| GPT-5.4 nano | OpenAI | $0.013 | $12.50 |
| Gemini 3.1 Flash-Lite | Google Gemini | $0.015 | $15.50 |
| Codestral | Mistral | $0.017 | $16.80 |
Try your own numbers in the token calculator, or compare coding tools on AI coding tools pricing.
Cheapest model from each provider
- OpenAI: GPT-6 Luna, $0.10 in and $0.50 out per 1M tokens.
- Anthropic Claude: Claude Haiku 5.5, $0.10 in and $0.50 out per 1M tokens.
- Google Gemini: Gemini 3.1 Flash-Lite, $0.25 in and $1.50 out per 1M tokens.
- DeepSeek: DeepSeek V4.1 Flash, $0.30 in and $1.20 out per 1M tokens.
- Mistral: Ministral 3 3B, $0.10 in and $0.10 out per 1M tokens.
Free usage: Google's Flash and Flash-Lite models have a free tier, and new Mistral accounts start in a free mode with monthly usage. See free tier limits.
FAQ
What is the cheapest LLM API?
On our November 1, 2026 check, the cheapest model by input plus output price was Ministral 3 3B at $0.10 in and $0.10 out per 1M tokens, then Ministral 3 8B and Ministral 3 14B. Google's Flash and Flash-Lite models and Mistral's free mode also offer free usage within limits.
What is the cheapest LLM API for coding?
By price alone, for a coding agent step of 50,000 input and 2,000 output tokens with no cache, the cheapest model on our list was Ministral 3 3B at about $0.0052 a step. This ranks cost only, not code quality; test a few models on your own tasks before you switch.
Is DeepSeek cheaper off-peak?
Yes. DeepSeek charges half price outside its peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays). Our tables use the peak price, so DeepSeek can cost half of what is shown.
How can I make any LLM API cheaper?
Keep a stable prompt prefix so it is served from the cache, send work that can wait through a batch API (about half price at OpenAI, Anthropic, Google and Mistral), and cap output length, since output tokens usually cost several times more than input.