Open-Weight Token Report: October 2026
Tokens served on open-weight models through a large public LLM router in September 2026, by model family and provider.
What moved in September
Open-weight volume rose 57% a day. 368T open-weight tokens were served through the router in September 2026, 12.3T a day against 7.8T in August. The router is one channel among many: direct and enterprise traffic is not included.
DeepSeek was the largest family, with 33%. GLM gained the most share, +10.2 points to 21%.
Relace added the most volume among providers serving other labs' models. Its open-weight tokens rose from 162B a day in August to 818B in September. Next: Together (+636B a day) and Wafer (+276B a day).
Daily volume by model family
- DeepSeek
- GLM
- Hunyuan
- MiMo
- Nemotron
- MiniMax
- Other families
Model families
| Family | Lab | Tokens, September | Share | Share chg, pts | Per day chg | Top provider | Lab serves |
|---|---|---|---|---|---|---|---|
| DeepSeek | DeepSeek | 123T | 33.3% | -1.6 | +50% | Relace 13% | 13% |
| GLM | Z.ai (Zhipu) | 76.4T | 20.8% | +10.2 | +208% | Z.ai 20% | 20% |
| Hunyuan | Tencent | 71.5T | 19.4% | +3.4 | +90% | Tencent Cloud 99% | 99% |
| MiMo | Xiaomi | 31.9T | 8.7% | -4.4 | +4% | Xiaomi 99% | 99% |
| Nemotron | NVIDIA | 24.1T | 6.6% | -2.0 | +20% | NVIDIA 99% | 99% |
| MiniMax | MiniMax | 12.0T | 3.3% | -1.3 | +13% | GMICloud 51% | 42% |
| Other open-weight | Various | 10.8T | 2.9% | -1.4 | +6% | Poolside 52% | n/a |
| Kimi | Moonshot AI | 7.6T | 2.1% | -0.9 | +9% | Moonshot AI 19% | 19% |
| Qwen | Alibaba (Qwen team) | 3.2T | 0.9% | +0.3 | +150% | Alibaba Cloud Int. 51% | 51% |
| gpt-oss | OpenAI | 3.1T | 0.8% | -0.3 | +19% | Groq 21% | n/a |
| Gemma | 2.7T | 0.7% | -0.5 | -8% | DeepInfra 24% | n/a | |
| Mistral | Mistral AI | 1.2T | 0.3% | 0.0 | +60% | DeepInfra 81% | 10% |
| Step | StepFun | 515B | 0.1% | -1.4 | -85% | StepFun 100% | 100% |
| Llama | Meta | 233B | 0.1% | -0.1 | -30% | Groq 80% | n/a |
Lab serves: share of the family's tokens served by the lab that made it; n/a where the lab sells no API through the router.
Provider leaderboard
| # | Provider | Tokens, September | Share | Share, August | Per day chg | Main family |
|---|---|---|---|---|---|---|
| 1 | Tencent Cloudlab | 71.1T | 19.3% | 15.8% | +92% | Hunyuan |
| 2 | Xiaomilab | 31.5T | 8.6% | 13.0% | +4% | MiMo |
| 3 | Relace | 24.5T | 6.7% | 2.1% | +404% | DeepSeek |
| 4 | NVIDIAlab | 23.9T | 6.5% | 8.4% | +21% | Nemotron |
| 5 | Together | 21.4T | 5.8% | 1.0% | +843% | DeepSeek |
| 6 | NovitaAI | 17.4T | 4.7% | 7.4% | 0% | DeepSeek |
| 7 | DeepSeeklab | 15.5T | 4.2% | 4.9% | +35% | DeepSeek |
| 8 | DeepInfra | 15.4T | 4.2% | 6.4% | +3% | DeepSeek |
| 9 | Z.ailab | 15.2T | 4.1% | 3.5% | +84% | GLM |
| 10 | StreamLake | 15.1T | 4.1% | 4.1% | +57% | DeepSeek |
| 11 | GMICloud | 13.9T | 3.8% | 4.5% | +32% | MiniMax |
| 12 | CoreWeave | 9.4T | 2.6% | 2.7% | +47% | GLM |
| 13 | Wafer | 8.9T | 2.4% | 0.3% | +1420% | DeepSeek |
| 14 | Baidu Qianfan | 8.2T | 2.2% | 4.5% | -23% | DeepSeek |
| 15 | Parasail | 7.0T | 1.9% | 0.7% | +337% | GLM |
Share of all open-weight tokens. Lab: a lab serving its own models.
Fastest-growing providers
| Provider | Per day, August | Per day, September | Added per day | Change | Last 7 days of September |
|---|---|---|---|---|---|
| Tencent Cloudlab | 1.2T | 2.4T | +1.1T | +92% | 1.4T |
| Relace | 162B | 818B | +656B | +404% | 878B |
| Together | 76B | 712B | +636B | +843% | 1.4T |
| Wafer | 19B | 296B | +276B | +1420% | 405B |
| Z.ailab | 275B | 506B | +230B | +84% | 155B |
| Open Inference | 21B | 218B | +197B | +934% | 256B |
| StreamLake | 320B | 504B | +184B | +57% | 355B |
| Parasail | 54B | 234B | +181B | +337% | 335B |
| NVIDIAlab | 659B | 795B | +137B | +21% | 988B |
| DeepSeeklab | 382B | 515B | +133B | +35% | 510B |
Ranked by open-weight tokens added a day, among providers serving at least 20B a day in September.
Other editions
- September 2026 edition (August 2026 data, 242T open-weight tokens)
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