Open-Weight LLM Price Index
What a million tokens of an open-weight model costs, across the providers that sell it. Daily list prices, July 12 to October 9, 2026.
What moved
The cheapest output prices fell 35% in 30 days; the median fell 6%. Weighted by each model's tokens, for models listed all 30 days. These are list prices through a large public LLM router, net of listed discounts: one per provider, its cheapest paid endpoint at the end of each UTC day. The router is one channel among many: direct and enterprise prices are not included.
Same weights, very different prices. On DeepSeek V4.1 Flash, the most-used model in the index, 29 providers list output tokens from $0.18 to $1.60 per million, around a median of $1.00.
Price by model family
| Family | Model | Providers | Output, median | Output, cheapest | Input, median | Spread | 30 days | 90 days |
|---|---|---|---|---|---|---|---|---|
| DeepSeek | DeepSeek V4.1 Flash +6 more | 29 | $1.00 | $0.18 | $0.18 | 8.9x | -17% | new |
| GLM | GLM 5.3 Flash +5 more | 29 | $0.50 | $0.25 | $0.113 | 6.4x | 0% | new |
| MiMo | MiMo-V2.6-Flash +2 more | 8 | $0.28 | $0.28 | $0.14 | 1.6x | 0% | new |
| Hunyuan | Hy3 | 5 | $0.58 | $0.33 | $0.14 | 2.4x | 0% | -16% |
| Kimi | Kimi K3 +3 more | 20 | $14.00 | $12.75 | $2.65 | 1.4x | -7% | new |
| MiniMax | MiniMax M3 +2 more | 13 | $1.20 | $0.96 | $0.30 | 3.1x | 0% | 0% |
| Nemotron | Nemotron 3.5 Lightning | 6 | $0.147 | $0.12 | $0.053 | 1.7x | -35% | new |
| Qwen | Qwen3.8 27B +8 more | 19 | $2.30 | $1.35 | $0.225 | 3.5x | -10% | new |
| gpt-oss | gpt-oss-120b +1 more | 20 | $0.55 | $0.17 | $0.10 | 5.6x | -8% | -8% |
| Gemma | Gemma 4 31B +2 more | 12 | $0.40 | $0.34 | $0.14 | 3.4x | 0% | 0% |
| Mistral | Mistral Nemo | 5 | $0.03 | $0.024 | $0.029 | 6.3x | -68% | -25% |
| Llama | Llama 3.1 8B Instruct +1 more | 5 | $0.08 | $0.04 | $0.05 | 7.2x | 0% | +23% |
Each family's most-used model. USD per million tokens. Spread: most expensive over cheapest listing. 30 and 90 days: change in the median output price; new means first listed after July 12, 2026.
All indexed models
| Model | Family | Providers | Output, median | Middle half | Cheapest (provider) | Input, median | 30 days |
|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek | 29 | $1.00 | $0.60-$1.20 | $0.18 (Decart) | $0.18 | -17% |
| GLM 5.3 Flash | GLM | 29 | $0.50 | $0.50-$0.50 | $0.23 (StreamLake) | $0.113 | 0% |
| MiMo-V2.6-Flash | MiMo | 8 | $0.28 | $0.28-$0.297 | $0.28 (Xiaomi) | $0.14 | 0% |
| DeepSeek V4 Flash 0731 | DeepSeek | 24 | $0.528 | $0.276-$1.00 | $0.18 (DeepInfra) | $0.135 | +51% |
| GLM 5.3 | GLM | 31 | $4.40 | $3.46-$4.40 | $2.20 (NovitaAI) | $1.12 | 0% |
| DeepSeek V4 Flash 0423 | DeepSeek | 16 | $0.28 | $0.19-$0.515 | $0.165 (StreamLake) | $0.114 | 0% |
| Hy3 | Hunyuan | 5 | $0.58 | $0.58-$0.64 | $0.33 (Tencent Cloud) | $0.14 | 0% |
| Kimi K3 | Kimi | 20 | $14.00 | $13.38-$15.00 | $9.60 (Decart) | $2.65 | -7% |
| GLM 5.2 | GLM | 25 | $4.40 | $2.95-$4.40 | $1.68 (Decart) | $1.19 | +5% |
| MiniMax M3 | MiniMax | 13 | $1.20 | $1.20-$1.20 | $0.96 (GMICloud) | $0.30 | 0% |
| DeepSeek V4 Pro 0813 | DeepSeek | 19 | $3.96 | $2.90-$3.96 | $0.657 (Baidu Qianfan) | $1.30 | 0% |
| Nemotron 3.5 Lightning | Nemotron | 6 | $0.147 | $0.131-$0.19 | $0.12 (Darkbloom) | $0.053 | -35% |
| MiMo-V2.5 | MiMo | 5 | $0.336 | $0.28-$0.336 | $0.238 (GMICloud) | $0.168 | +9% |
| DeepSeek V4 Pro 0423 | DeepSeek | 15 | $3.20 | $2.58-$3.43 | $0.375 (Baidu Qianfan) | $1.30 | +1% |
| Qwen3.8 27B | Qwen | 19 | $2.30 | $1.93-$3.00 | $1.35 (NEAR AI) | $0.225 | -10% |
| gpt-oss-120b | gpt-oss | 20 | $0.55 | $0.25-$0.60 | $0.15 (Venice) | $0.10 | -8% |
| Gemma 4 31B | Gemma | 12 | $0.40 | $0.367-$1.00 | $0.34 (DeepInfra) | $0.14 | 0% |
| Gemma 4 26B A4B | Gemma | 13 | $0.33 | $0.30-$0.40 | $0.21 (io.net) | $0.10 | -18% |
| DeepSeek V3.2 | DeepSeek | 12 | $0.69 | $0.388-$1.54 | $0.31 (GMICloud) | $0.29 | +8% |
| gpt-oss-20b | gpt-oss | 10 | $0.145 | $0.133-$0.172 | $0.09 (Darkbloom) | $0.03 | -3% |
| Mistral Nemo | Mistral | 5 | $0.03 | $0.03-$0.03 | $0.024 (io.net) | $0.029 | -68% |
| MiMo-V2.5-Pro | MiMo | 6 | $0.915 | $0.87-$1.02 | $0.609 (GMICloud) | $0.458 | 0% |
| Kimi K2.6 | Kimi | 17 | $3.50 | $3.40-$4.00 | $1.83 (Baidu Qianfan) | $0.75 | 0% |
| Qwen3 235B A22B Instruct 2507 | Qwen | 6 | $0.675 | $0.563-$0.787 | $0.35 (GMICloud) | $0.145 | 0% |
| Qwen3.8 2.4T A95B | Qwen | 7 | $6.00 | $6.00-$6.00 | $6.00 (SiliconFlow) | $2.00 | 0% |
| Qwen3.6 35B A3B | Qwen | 10 | $1.00 | $0.963-$1.22 | $0.70 (Darkbloom) | $0.125 | 0% |
| MiniMax M2.7 | MiniMax | 5 | $1.20 | $1.08-$1.20 | $0.84 (GMICloud) | $0.30 | 0% |
| Gemma 3 27B | Gemma | 5 | $0.30 | $0.20-$0.30 | $0.16 (DeepInfra) | $0.08 | +20% |
| GLM 4.7 | GLM | 6 | $2.09 | $1.94-$2.20 | $1.75 (DeepInfra) | $0.57 | -5% |
| DeepSeek V3.1 | DeepSeek | 5 | $1.50 | $1.00-$1.65 | $0.95 (DeepInfra) | $0.55 | 0% |
| GLM 5 | GLM | 8 | $2.88 | $2.16-$3.20 | $1.92 (GMICloud) | $0.975 | 0% |
| Kimi K2.7 Code | Kimi | 11 | $3.84 | $3.50-$4.00 | $3.00 (StreamLake) | $0.912 | 0% |
| Llama 3.1 8B Instruct | Llama | 5 | $0.08 | $0.05-$0.22 | $0.04 (DeepInfra) | $0.05 | 0% |
| Kimi K2.5 | Kimi | 5 | $2.85 | $2.50-$3.00 | $2.50 (SiliconFlow) | $0.532 | 0% |
| Llama 3.3 70B Instruct | Llama | 10 | $0.715 | $0.505-$0.873 | $0.32 (DeepInfra) | $0.371 | 0% |
| Qwen3.5-9B | Qwen | 6 | $0.15 | $0.15-$0.225 | $0.13 (Darkbloom) | $0.10 | 0% |
| Qwen3.5 397B A17B | Qwen | 10 | $3.55 | $3.50-$3.60 | $2.34 (Alibaba Cloud Int.) | $0.55 | 0% |
| GLM 5.1 | GLM | 13 | $4.40 | $3.96-$4.40 | $3.04 (StreamLake) | $1.38 | 0% |
| Qwen3.6 27B | Qwen | 7 | $3.20 | $2.35-$3.23 | $2.00 (Chutes) | $0.32 | +8% |
| MiniMax M2.5 | MiniMax | 8 | $1.20 | $1.17-$1.20 | $0.95 (Venice) | $0.30 | 0% |
| Qwen3.5-35B-A3B | Qwen | 7 | $1.00 | $1.00-$1.55 | $0.75 (Darkbloom) | $0.15 | -20% |
| Qwen3.5-27B | Qwen | 6 | $2.28 | $2.04-$2.40 | $1.56 (Alibaba Cloud Int.) | $0.265 | 0% |
Models with at least five paid providers, most-used first. Prices as of October 9, 2026.
Discuss a transaction
Serving open-weight models on your own GPUs, or lending to a provider that does? Tell us about the cluster and the financing you need.
Prefer email? hello@amcompute.com