ModelRun [by Modular]: Models and Prices
Modular Cloud, the hosted inference service of Modular (maker of Mojo and MAX): shared OpenAI-compatible endpoints billed per token and dedicated deployments.
ModelRun [by Modular] served 551B tokens of other labs' open-weight models over Nov 2 to Nov 8, 2026 through a large public LLM router, #28 among providers.
Open-weight models
| Model | Family | Share of model | Tokens, 7d | Input $/M | Output $/M | Tokens/s | Uptime 3d | Context |
|---|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | DeepSeek | 0.6% | 319B | $0.17 | $0.67 | 117 | 100.0% | 1.05M |
| MiniMax M3.1 Flash | MiniMax | 1.8% | 91B | $0.15 | $0.60 | 116 | 100.0% | 1.05M |
| Qwen3.8 27B | Qwen | 11.2% | 66B | $0.20 | $2.50 | 117 | 100.0% | 262K |
| MiniMax M3 | MiniMax | 5.9% | 35B | $0.30 | $1.30 | 87 | 100.0% | 1.05M |
| Gemma 4 31B | Gemma | 8.0% | 34B | $0.75 | $1.00 | 215 | 100.0% | 262K |
List prices per million tokens as of November 8, 2026. Share: of all tokens served on that model; a dash where the volume is not reported.
Company
- Website
- modular.com
- Headquarters
- Los Altos, CA
- Compute
- Serves production workloads on NVIDIA and AMD GPUs; dedicated deployments run on Modular’s compute or the customer’s own cloud.
- Latest financing
- Acquired by Qualcomm; the deal closed in July 2026.
Finance your GPUs
Serving tokens on your own GPUs? Tell us the cluster, the models you serve and the contracts behind them, and we will share it with partner funders that fit.
Prefer email? hello@amcompute.com