Quick answer
The cheapest models are GPT-6 Luna and Claude Haiku 5.5, both $0.10 input and $0.50 output per million tokens. The cheapest open-weight model is Tencent Hy3 at $0.132 / $0.528. DeepSeek V4.1 Flash drops to $0.15 / $0.60 off-peak.
All 25 models, cheapest first
Ranked by output price, since output usually costs four to five times more than input.
| Model | Context | Input $/1M | Output $/1M | Best for |
|---|---|---|---|---|
GPT-6 Luna OpenAI, US | 1.05M | $0.10 | $0.50 | High-volume tasks; runs ChatGPT Free and Go |
Claude Haiku 5.5 Anthropic, US | 1M | $0.10 | $0.50 | Fast, cheap, high volume |
Hy3 Tencent, China | 256K | $0.132 | $0.528 | Cheapest open-weight model here (Apache-2.0) |
MiMo-v2.6 Pro Xiaomi, China | Not stated | $0.435 | $0.87 | Very low-cost reasoning model from Xiaomi |
DeepSeek V4.1 Flash DeepSeek, China | 1M | $0.30 | $1.20 | Fast open model; half price off-peak ($0.15 / $0.60) |
MiniMax M3 MiniMax, China | 1M | $0.30 | $1.20 | Low-cost model that reads images and video; community license |
Mistral Large 4 Mistral, Europe | 1M | $0.70 | $2.10 | Mistral’s flagship, with open weights under Mistral’s license |
Doubao Seed 2.0 Pro ByteDance, China | 256K | $0.50 | $3.00 | Model behind China’s most-used chatbot, Doubao |
DeepSeek V4 Pro DeepSeek, China | 1M | $1.32 | $3.96 | Strong open model; half price off-peak ($0.66 / $1.98) |
Muse Spark 1.3 Meta, US | 1M | $1.25 | $4.25 | Multimodal reasoning at a lower price |
GLM-5.3 Zhipu, China | 1M | $1.40 | $4.40 | Zhipu’s newest flagship, strong at coding |
Qwen4 Max Alibaba, China | 1M | $1.50 | $4.50 | Alibaba’s newest flagship, now on pay-as-you-go |
GPT-6.1 Terra OpenAI, US | 1.05M | $1.00 | $5.00 | Fast, efficient everyday work |
Grok 4.7 xAI, US | 500K | $2.00 | $6.00 | Coding, agents and knowledge work |
Gemini 4 Hydrogen Google, US | 1.05M | $1.50 | $7.50 | Everyday coding and agent workflows; free tier |
Mistral Medium 3.5 Mistral, Europe | 256K | $1.50 | $7.50 | Being retired in April 2027; Large 4 replaces it |
GPT-6.1 Sol OpenAI, US | 1.05M | $2.00 | $10.00 | Near-Astra quality at a fifth of the price |
Claude Sonnet 5.5 Anthropic, US | 1M | $2.00 | $10.00 | Balanced quality and cost |
Gemini 4 Carbon Google, US | 1.05M | $2.00 | $10.00 | Google’s model for the hardest reasoning problems |
Gemini 4 Argon Google, US | 1.05M | $2.00 | $10.00 | Deep reasoning with faster replies than Carbon |
Kimi K3.1 Pro Moonshot, China | 1M | $2.00 | $10.00 | Coding and agents; weights released Nov 5 |
Claude Opus 5.5 Anthropic, US | 1M | $4.00 | $20.00 | Long, complex tasks and agents |
Muse Wonder Meta, US | 1M | $4.00 | $20.00 | Meta’s frontier model for agents |
Claude Fable 5.2 Anthropic, US | 1M | $8.00 | $40.00 | Anthropic’s top model for coding and knowledge work |
GPT-6 Astra OpenAI, US | 1.05M | $10.00 | $50.00 | OpenAI’s most capable model, for the hardest reasoning |
Ways to pay less
- OpenAI’s Batch and Flex requests and Mistral’s batch processing cost 50% of standard rates.
- DeepSeek charges half price outside peak hours (peak is 01:00–04:00 and 06:00–10:00 UTC on weekdays).
- Gemini 4 Argon and Carbon cost $2 / $10 until November 30, 2026, then $4 / $20.
- Some models charge more for very long prompts: Grok 4.7 above 200K tokens, Claude Haiku 5.5 above 100K, Doubao Seed 2.0 Pro above 128K and MiniMax M3 above 512K.
Questions people ask
What is the cheapest AI API?
GPT-6 Luna and Claude Haiku 5.5, at $0.10 per million input tokens and $0.50 per million output tokens. Both have a 1M-token context window.
What is the cheapest open-weight AI model?
Tencent Hy3, at $0.132 input and $0.528 output per million tokens on Tencent Cloud, with Apache-2.0 weights you can run yourself.
How can I cut my AI API costs further?
Use batch processing where offered (OpenAI and Mistral charge 50% less), cache repeated prompts, and run DeepSeek jobs off-peak at half price.