gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized to run on a single H100 GPU with native MXFP4 quantization. The model supports configurable reasoning depth, full chain-of-thought access, and native tool use, including function calling, browsing, and structured output generation.
| $0.03 | $0.15 | $0.015 | 0.63s | 47 tps | ||
| $0.03 | $0.15 | $0.03 | 0.77s | 44 tps | ||
| $0.03 | $0.17 | $0.03 | 0.31s | 108 tps | ||
| $0.03 | $0.18 | $0.03 | 0.46s | 81 tps | ||
| $0.03 | $0.187 | $0.03 | 0.32s | 140 tps | ||
| $0.031 | $0.116 | $0.028 | 0.62s | 75 tps | ||
| $0.037 | $0.17 | -- | 0.47s | 32 tps | ||
| $0.05 | $0.25 | $0.05 | 0.24s | 245 tps | ||
| $0.05 | $0.25 | -- | 0.68s | 175 tps | ||
| $0.055 | $0.25 | -- | 0.85s | 47 tps | ||
| $0.06 | $0.40 | $0.03 | 0.59s | 151 tps | ||
| $0.06 | $0.42 | $0.012 | 0.62s | 51 tps | ||
| $0.10 | $0.50 | $0.10 | 0.16s | 188 tps | ||
| $0.10 | $0.50 | $0.10 | 0.19s | 191 tps | ||
| $0.10 | $0.75 | $0.055 | 0.37s | 175 tps | ||
| $0.14 | $0.95 | -- | 0.88s | 399 tps | ||
| $0.15 | $0.60 | $0.10 | 0.27s | 316 tps | ||
| $0.15 | $0.60 | -- | 0.54s | 333 tps | ||
| $0.15 | $0.60 | $0.075 | 1.14s | 36 tps | ||
| $0.15 | $0.60 | -- | 0.81s | 168 tps | ||
| $0.15 | $0.60 | $0.075 | 0.18s | 335 tps | ||
| $0.15 | $0.75 | -- | 0.90s | 214 tps | ||
| $0.35 | $0.75 | $0.35 | 0.16s | 809 tps | ||
| $0.09 | $0.36 | -- | 0.33s | 49 tps | ||
| $0.15 | $0.60 | -- | 0.24s | 147 tps | ||
| $0.20 | $0.95 | -- | 1.55s | 56 tps | ||
Standard ルーティングでは使用されません:これらのエンドポイントが使用されない理由 | ||||||
| $0.15 | $0.60 | -- | 0.41s | 181 tps | ||
P50、プロバイダー全体での最良値
P50、最適なプロバイダー
上流のプロバイダーでエラーが発生した場合、リクエストフィルターが許可していれば、別の正常なプロバイダーにルーティングして復旧できます。プロバイダーごとの稼働率データにプログラムでアクセスするには Endpoints APIを利用できます。詳しくは、 負荷分散とプロバイダールーティングのオプションをご覧ください。