of gpt-oss-120b's tool calls fail on NovitaAI (fp4).
Updated October 29, 2026
Tool-call index
Not all models are served equally by all providers. On a typical model, the worst provider breaks tool calls 19× as often as the best. Find the best one for your agents.
| # | Model | Broken tool calls | Rank | Best on the model | Tool calls a day |
|---|---|---|---|---|---|
| 1 | Kimi K2.7 Codeint4 | 0.03% | 1 of 11 | — | 5.9k |
| 2 | Kimi K2.6 | 0.13% | 2 of 14 | SiliconFlow (fp8) 0.06% | 9.4k |
| 3 | Hy4 previewfp8 | 0.15% | 2 of 6 | DeepInfra (fp4) 0.10% | 238k |
| 4 | DeepSeek V4 Pro 0813fp8 | 0.47% | 5 of 21 | Ionstream 0.04% | 50k |
| 5 | Qwen3.5 397B A17B | 0.72% | 2 of 7 | Alibaba Cloud Int. 0.58% | 2.8k |
| 6 | GLM 5.2fp8 | 0.83% | 12 of 27 | Friendli 0.15% | 24k |
| 7 | MiniMax M3fp8 | 0.92% | 3 of 11 | ModelRun [by Modular] (fp8) 0.27% | 22k |
| 8 | Gemma 4 31Bbf16 | 0.96% | 3 of 12 | ModelRun [by Modular] (fp4) 0.09% | 5.1k |
| 9 | Gemma 4 26B A4Bbf16 | 1.02% | 4 of 11 | NextBit (bf16) 0.00% | 15k |
| 10 | GLM 4.7fp8 | 1.04% | 2 of 3 | Z.ai (fp4) 0.93% | 2.0k |
| 11 | DeepSeek V4 Flash 0731fp8 | 1.27% | 7 of 22 | Sail Research (fp4) 0.04% | 106k |
| 12 | DeepSeek V4 Pro 0423fp8 | 1.28% | 4 of 13 | Cloudflare 0.78% | 138k |
| 13 | MiniMax M3.1 Flashfp16 | 2.05% | 6 of 11 | Together 1.17% | 17k |
| 14 | GLM 5.3 Flashfp8 | 2.26% | 30 of 34 | NextBit (fp4) 0.00% | 404k |
| 15 | MiMo-V2.6-Flashfp8 | 2.82% | 6 of 9 | Venice (fp8) 1.41% | 76k |
| 16 | Qwen3.8 2.4T A95B | 3.11% | 4 of 5 | SiliconFlow (fp8) 0.19% | 4.5k |
| 17 | Hy3 | 3.23% | 5 of 5 | Tencent Cloud (fp8) 1.46% | 30k |
| 18 | DeepSeek V4.1 Flash 1017fp8 | 3.50% | 33 of 35 | Sail Research (fp4) 0.03% | 131k |
| 19 | Kimi K2.5 | 3.61% | 4 of 5 | SiliconFlow (int4) 0.17% | 4.1k |
| 20 | MiniMax M2.7fp8 | 3.94% | 6 of 6 | Groq 0.32% | 9.9k |
| 21 | Qwen3.8 27B | 4.41% | 20 of 21 | Darkbloom (fp4) 0.07% | 14k |
| 22 | DeepSeek V4.1 Flashfp8 | 5.20% | 35 of 35 | Prime Intellect 0.06% | 331k |
| 23 | MiMo-V2.6-Profp8 | 5.47% | 4 of 4 | DeepInfra (fp8) 1.48% | 343k |
| 24 | GLM 5.3fp8 | 6.71% | 41 of 42 | Amazon Bedrock 0.08% | 41k |
| 25 | Llama 3.3 70B Instructbf16 | 25.58% | 4 of 5 | Groq 2.21% | 2.4k |
| 26 | gpt-oss-120bfp4 | 29.28% | 13 of 15 | Groq 0.04% | 13k |
Get updates when a new model drops
And which provider to use.
For inference providers
Tool calling doesn't have to break
The same provider can handle tool calls well on one model and struggle on the next.
on Kimi K2.7 Code, its best model.
models, where the settings tell OpenRouter to skip NovitaAI.