Inference Uptime Tracker
Open-weight inference providers ranked by uptime, October 8 to 10, 2026.
Uptime by provider
5 of 51 providers held 99.9% uptime or better; 26 fell below 99%. Uptime is the share of measured minutes in which a provider's endpoints served requests through a large public LLM router. The router publishes only the last three days, so read this as a snapshot. It is one channel among many: direct and enterprise traffic is not included.
| # | Provider | Uptime | Token-weighted | Downtime a month | Worst model-day | Tokens, 7 days |
|---|---|---|---|---|---|---|
| 1 | Relace | 99.99% | 100.00% | 5 min | 99.8%, DeepSeek V4 Pro 0423, Oct 8 | 8.91T |
| 2 | Moonshot AI | 99.97% | 100.00% | 13 min | 99.8%, Kimi K2.6, Oct 9 | 312B |
| 3 | inference.net | 99.96% | 99.98% | 18 min | 97.8%, Schematron V2 Turbo, Oct 9 | 4.09T |
| 4 | Tencent Cloud | 99.95% | 99.90% | 21 min | 99.7%, Hy4 preview, Oct 8 | 9.71T |
| 5 | Modal | 99.94% | 99.99% | 28 min | 97.7%, Qwen3.8 2.4T A95B, Oct 8 | 460B |
| 6 | MiniMax | 99.88% | 99.95% | 53 min | 96.2%, MiniMax M2.5, Oct 8 | 1.07T |
| 7 | ModelRun [by Modular] | 99.83% | 99.91% | 1 h 14 min | 99.0%, Gemma 4 31B, Oct 8 | 283B |
| 8 | CoreWeave | 99.82% | 99.75% | 1 h 19 min | 97.2%, Qwen3.8 27B, Oct 9 | 1.55T |
| 9 | Baseten | 99.81% | 99.69% | 1 h 22 min | 98.4%, gpt-oss-120b, Oct 10 | 598B |
| 10 | AkashML | 99.78% | 99.83% | 1 h 34 min | 98.7%, Qwen3.6 35B A3B, Oct 8 | 89B |
| 10 | Amazon Bedrock | 99.78% | 99.34% | 1 h 37 min | 98.3%, Kimi K3, Oct 9 | 65B |
| 12 | Wafer | 99.74% | 99.93% | 1 h 52 min | 96.3%, Kimi K3, Oct 8 | 2.57T |
| 13 | Darkbloom | 99.64% | 99.96% | 2 h 36 min | 96.8%, Qwen3.8 27B, Oct 10 | 108B |
| 14 | Cloudflare | 99.62% | 99.56% | 2 h 44 min | 95.5%, GLM 5.2, Oct 10 | 76B |
| 15 | Z.ai | 99.59% | 100.00% | 2 h 59 min | 89.5%, GLM 4.5 Air, Oct 8 | 1.30T |
| 16 | DekaLLM | 99.50% | 99.63% | 3 h 36 min | 96.8%, GLM 5.3 Flash, Oct 9 | 163B |
| 17 | Friendli | 99.49% | 99.19% | 3 h 40 min | 96.3%, Gemma 4 31B, Oct 8 | 350B |
| 18 | Mistral | 99.46% | 99.68% | 3 h 53 min | 96.5%, Ministral 3 14B 2512, Oct 8 | 600B |
| 19 | Decart | 99.44% | 99.86% | 4 h 1 min | 96.3%, Kimi K3, Oct 10 | 973B |
| 20 | Makora | 99.31% | 99.61% | 4 h 57 min | 96.5%, GLM 5.3, Oct 8 | 414B |
| 20 | Sail Research | 99.31% | 98.68% | 4 h 58 min | 92.4%, DeepSeek V4 Flash 0731, Oct 8 | 1.71T |
| 22 | io.net | 99.24% | 99.65% | 5 h 30 min | 92.7%, Gemma 4 31B, Oct 8 | 97B |
| 23 | Parasail | 99.19% | 99.90% | 5 h 51 min | 43.0%, Qwen3 VL 8B Instruct, Oct 10 | 2.67T |
| 24 | Inceptron | 99.17% | 99.28% | 5 h 57 min | 94.9%, GLM 5.3 Flash, Oct 8 | 160B |
| 25 | Alibaba Cloud Int. | 99.11% | 99.60% | 6 h 22 min | 88.6%, Qwen3.5-122B-A10B, Oct 10 | 1.01T |
| 26 | DigitalOcean | 98.90% | 99.74% | 7 h 55 min | 68.6%, Qwen3.5 397B A17B, Oct 10 | 1.37T |
| 27 | Crusoe | 98.51% | 99.29% | 10 h 46 min | 95.3%, gpt-oss-120b, Oct 9 | 202B |
| 28 | DeepInfra | 98.49% | 99.97% | 10 h 50 min | 70.1%, DeepSeek V3, Oct 8 | 6.43T |
| 29 | Baidu Qianfan | 98.22% | 98.48% | 12 h 48 min | 79.1%, DeepSeek V4 Pro 0423, Oct 8 | 500B |
| 30 | Xiaomi | 98.12% | 99.54% | 13 h 32 min | 93.1%, MiMo-V2.5, Oct 10 | 11.8T |
| 31 | Reka AI | 98.08% | 98.97% | 13 h 51 min | 87.3%, DeepSeek V4 Pro 0423, Oct 8 | 279B |
| 32 | Ionstream | 98.02% | 97.92% | 14 h 17 min | 90.3%, DeepSeek V4 Pro 0813, Oct 8 | 103B |
| 32 | NextBit | 98.02% | 99.83% | 14 h 17 min | 92.9%, Gemma 3 27B, Oct 9 | 78B |
| 34 | Morph | 97.57% | 97.97% | 17 h 31 min | 90.3%, GLM 5.3 Flash, Oct 8 | 1.02T |
| 35 | Phala | 97.52% | 98.44% | 17 h 51 min | 70.3%, Qwen3.8 27B, Oct 10 | 311B |
| 36 | Together | 97.51% | 99.69% | 17 h 54 min | 82.2%, Inkling, Oct 8 | 11.5T |
| 37 | AtlasCloud | 97.10% | 99.37% | 20 h 54 min | 49.5%, Qwen3.5-122B-A10B, Oct 10 | 1.17T |
| 38 | Open Inference | 97.05% | 96.46% | 21 h 13 min | 82.2%, DeepSeek V4.1 Flash, Oct 10 | 759B |
| 39 | Mancer | 97.04% | 98.23% | 21 h 20 min | 70.1%, GLM 4.7, Oct 9 | 17B |
| 40 | Venice | 96.92% | 99.22% | 22 h 12 min | 69.8%, Qwen3 Coder 480B A35B, Oct 8 | 465B |
| 41 | SiliconFlow | 96.86% | 97.59% | 22 h 35 min | 68.7%, gpt-oss-120b, Oct 9 | 893B |
| 42 | GMICloud | 96.77% | 97.30% | 23 h 16 min | 64.5%, MiMo-V2.6-Flash, Oct 10 | 1.45T |
| 43 | MARA | 96.49% | 96.99% | 25 h 18 min | 55.0%, DeepSeek V3.2, Oct 8 | 5B |
| 44 | Fireworks | 95.81% | 89.81% | 30 h 12 min | 39.9%, DeepSeek V4.1 Flash, Oct 10 | 2.20T |
| 45 | StreamLake | 95.59% | 99.73% | 31 h 45 min | 8.2%, MiMo-V2.5, Oct 10 | 2.52T |
| 46 | NovitaAI | 95.47% | 98.29% | 32 h 36 min | 5.4%, GLM 4.7 Flash, Oct 10 | 3.04T |
| 47 | Nebius Token Factory | 94.93% | 93.43% | 36 h 30 min | 74.8%, GLM 5.2, Oct 8 | 34B |
| 47 | SambaNova | 94.93% | 95.98% | 36 h 30 min | 85.0%, Gemma 4 31B, Oct 10 | 8B |
| 49 | Groq | 93.60% | 92.12% | 46 h 4 min | 2.6%, gpt-oss-20b, Oct 10 | 161B |
| 50 | Chutes | 93.24% | 95.53% | 48 h 40 min | 68.3%, Gemma 4 31B, Oct 8 | 41B |
| 51 | Google Vertex | 90.42% | - | 69 h 1 min | 3.0%, gpt-oss-120b, Oct 9 | - |
Uptime is weighted by measured minutes; token-weighted weights each model by the provider's tokens on it. Downtime a month applies the three-day rate to 30 days. Ties share a rank.
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