Pricing · 2026-07-30
Cheapest AI models
The ten cheapest models that developers are genuinely running, by list price per million output tokens. Free demo endpoints and models with no measurable traffic are excluded.
What cheap costs in July 2026
Across 50 tracked models, gateway traffic of roughly 7.7T tokens a day implies about $23,939,776 of daily list-price spend. Most of that concentrates in a handful of flagships, which is exactly why the cheap tier matters: swapping a high-volume, low-complexity workload onto a model in this table changes the bill by an order of magnitude.
The cheapest model with real usage is inclusionAI: Ling-2.6-flash at $0.03 per million output tokens, holding 0.44% of tracked tokens (+0.08pp over 30 days).
Top 10 cheapest AI models in use
List price per million output tokens, models with real gateway traffic.
| # | Model | Lab | $/M out | $/M in | Share % | Tokens/d |
|---|---|---|---|---|---|---|
| 1 | inclusionAI: Ling-2.6-flash | Other | $0.03 | $0.01 | 0.44 | 36B |
| 2 | Poolside: Laguna XS 2.1 | Other | $0.12 | $0.06 | 0.34 | 28B |
| 3 | Qwen: Qwen3.7 Flash | Alibaba | $0.13 | $0.03 | 0.32 | 26B |
| 4 | OpenAI: gpt-oss-120b | OpenAI | $0.17 | $0.04 | 0.97 | 81B |
| 5 | Poolside: Laguna S 2.1 | Other | $0.18 | $0.09 | 1.43 | 119B |
| 6 | DeepSeek: DeepSeek V4 Flash | DeepSeek | $0.28 | $0.14 | 13.96 | 1.2T |
| 7 | Xiaomi: MiMo-V2.5 | Xiaomi | $0.28 | $0.14 | 8.68 | 722B |
| 8 | Google: Gemma 4 31B | $0.34 | $0.10 | 0.72 | 60B | |
| 9 | Google: Gemma 4 26B A4B | $0.34 | $0.07 | 0.67 | 56B | |
| 10 | Google: Gemini 2.5 Flash Lite | $0.40 | $0.10 | 1.12 | 93B |
Cheapest option per lab
The lowest-priced model with real usage from each lab.
| # | Model | Lab | $/M out | $/M in | Share % | Tokens/d |
|---|---|---|---|---|---|---|
| 1 | inclusionAI: Ling-2.6-flash | Other | $0.03 | $0.01 | 0.44 | 36B |
| 2 | Qwen: Qwen3.7 Flash | Alibaba | $0.13 | $0.03 | 0.32 | 26B |
| 3 | OpenAI: gpt-oss-120b | OpenAI | $0.17 | $0.04 | 0.97 | 81B |
| 4 | DeepSeek: DeepSeek V4 Flash | DeepSeek | $0.28 | $0.14 | 13.96 | 1.2T |
| 5 | Xiaomi: MiMo-V2.5 | Xiaomi | $0.28 | $0.14 | 8.68 | 722B |
| 6 | Google: Gemma 4 31B | $0.34 | $0.10 | 0.72 | 60B | |
| 7 | NVIDIA: Nemotron 3 Super | Nvidia | $0.40 | $0.08 | 0.59 | 49B |
| 8 | Tencent: Hy3 | Tencent | $0.53 | $0.13 | 7.36 | 612B |
| 9 | MiniMax: MiniMax M2.7 | MiniMax | $1.00 | $0.25 | 0.28 | 23B |
| 10 | StepFun: Step 3.7 Flash | StepFun | $1.15 | $0.20 | 3.39 | 282B |
Cheapest AI models FAQ
- What is the cheapest AI model that people actually use?
- inclusionAI: Ling-2.6-flash from Other at $0.03 per million output tokens, while still carrying 0.44% of tracked gateway traffic on 2026-07-30.
- What are the top 10 cheapest AI models?
- Ranked by output list price among models with real usage, the cheapest start with inclusionAI: Ling-2.6-flash, Poolside: Laguna XS 2.1, Qwen: Qwen3.7 Flash. The full top 10 is in the table above and is recomputed daily.
- Why exclude free and zero-traffic models?
- A $0 price on a model nobody routes traffic to is not a usable option. This page only ranks models with a real list price and measurable gateway usage, so the cheapest entries are ones you can actually deploy.
- Is the cheapest model the best value?
- Not always. Cheap models can need more tokens or more retries for the same task. Pair this page with the usage-per-dollar table on the best LLMs page to weigh price against adoption.
- How often do prices update?
- Prices and usage are re-read from the gateway catalogues every day. The figures on this page are from 2026-07-30.