of Gemma 4 31B's tool calls fail on SiliconFlow (fp8).
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.
| # | Provider | Broken tool calls | Tool calls a day | $ in / out per M | Tokens/s | Uptime |
|---|---|---|---|---|---|---|
| 1 | ModelRun [by Modular]fp4# | 0.09% | 30k | $0.75 / $1.00 | 161 | 100.0% |
| 2 | Friendli# | 0.80% | 8.6k | $0.14 / $0.40 | 148 | 100.0% |
| 3 | NovitaAIbf16# | 0.96% | 5.1k | $0.14 / $0.40 | 18 | 88.3% |
| 4 | DeepInfrafp4/turbo# | 1.17% | 110k | $0.14 / $0.51 | 44 | 100.0% |
| 5 | Chutesfp4# | 1.34% | 15k | $0.12 / $0.37 | 13 | 98.1% |
| 6 | CoreWeavefp4# | 1.70% | 209k | $0.10 / $0.34 | 36 | 99.9% |
| 7 | Crusoebf16# | 1.75% | 23k | $0.14 / $0.40 | 20 | 99.6% |
| 8 | Parasailfp8# | 2.23% | 23k | $0.15 / $0.40 | 21 | 99.8% |
| 9 | DeepInfrafp8# | 2.33% | 51k | $0.20 / $0.40 | 20 | 99.8% |
| 10 | Venicefp4# | 2.63% | 2.2k | $0.12 / $0.36 | 36 | 100.0% |
| 11 | SambaNova# | 5.05% | 3.8k | $0.38 / $1.15 | 148 | 96.6% |
| 12 | SiliconFlowfp8# | 5.88% | 2.7k | $0.75 / $1.00 | 8 | 99.4% |
Also ignored, under 1,000 tool calls a day: DekaLLM 15.60%.
{
"provider": {
"order": [
"modelrun/fp4",
"friendli",
"novita/bf16"
],
"ignore": [
"sambanova",
"siliconflow/fp8",
"dekallm"
]
}
}Point your agent at llms.txt
And which provider to use.
Releasing a model? Test it before launch.
of Gemma 4 31B's tool calls fail on SiliconFlow (fp8).
as often on SiliconFlow (fp8) as on ModelRun [by Modular] (fp4), its best provider.
broken tool calls a day across its 11 providers.