of DeepSeek V4.1 Flash 1017's structured outputs fail on SiliconFlow (fp8).
Updated October 31, 2026 · How we measure
Structured-output index
The same model breaks your pipeline's structured outputs 15× more often on some inference providers than on others. Find the best one for your pipelines.
Give this to your agent ↓
| # | Provider | Structured outputs a day | $ in / out per M | Tokens/s | Uptime | |
|---|---|---|---|---|---|---|
| 1 | Reka AI# | 0.40% | 177k | $0.06 / $0.60 | 106 | 97.3% |
| 2 | DeepInfrafp8/flex# | 0.65% | 132k | $0.08 / $0.24 | 61 | 86.3% |
| 3 | inference.netfp8# | 0.70% | 2.8M | $0.08 / $0.55 | 121 | 100.0% |
| 4 | Parasailfp8# | 0.70% | 121k | $0.30 / $1.20 | 174 | 100.0% |
| 5 | Makorafp8# | 0.81% | 37k | $0.15 / $1.20 | 168 | 99.9% |
| 6 | AtlasCloudfp8# | 0.81% | 35k | $0.20 / $0.78 | 119 | 99.8% |
| 7 | DeepInfrafp8# | 0.90% | 678k | $0.10 / $0.29 | 63 | 99.9% |
| 8 | Fireworks# | 0.91% | 20k | $0.30 / $1.20 | 172 | 98.4% |
| 9 | Ionstream# | 0.92% | 228k | $0.17 / $0.49 | 160 | 99.8% |
| 10 | Sail Researchfp4# | 0.93% | 351k | $0.10 / $0.15 | 66 | 100.0% |
| 11 | Prime Intellect# | 0.99% | 41k | $0.30 / $1.20 | 126 | 100.0% |
| 12 | Morphfp8# | 1.00% | 1.0M | $0.05 / $0.19 | 73 | 100.0% |
| 13 | Together# | 1.05% | 1.4M | $0.30 / $1.20 | 279 | 100.0% |
| 14 | DeepInfrafp8/priority# | 1.10% | 417k | $0.15 / $0.44 | 85 | 100.0% |
| 15 | Relacefp4# | 1.16% | 560k | $0.04 / $0.50 | 134 | 100.0% |
| 16 | Decartfp4# | 1.18% | 149k | $0.12 / $0.49 | — | 98.0% |
| 17 | Fireworks Spotspot# | 1.20% | 268k | $0.21 / $0.84 | — | 98.6% |
| 18 | Wafermxfp4# | 1.24% | 688k | $0.04 / $0.50 | 186 | 100.0% |
| 19 | DigitalOcean# | 1.26% | 284k | $0.18 / $0.72 | 193 | 100.0% |
| 20 | Venicefp8# | 1.45% | 2.7k | $0.30 / $1.20 | 139 | 100.0% |
| 21 | Nebius Token Factoryfp8# | 1.45% | 32k | $0.30 / $1.20 | 194 | 99.7% |
| 22 | Cloudflare# | 1.58% | 60k | $0.10 / $0.60 | 96 | 99.9% |
| 23 | Crusoefp8# | 1.63% | 22k | $0.29 / $1.20 | 167 | 100.0% |
| 24 | Modal# | 1.73% | 46k | $0.25 / $1.00 | 199 | 99.9% |
| 25 | Particle AI# | 1.98% | 17k | $0.10 / $1.00 | 126 | 100.0% |
| 26 | GMICloudfp8# | 2.10% | 9.5k | $0.18 / $0.72 | 133 | 99.3% |
| 27 | Baidu Qianfanfp8# | 2.27% | 9.0k | $0.15 / $0.60 | 108 | 100.0% |
| 28 | Cohere# | 3.01% | 69k | $0.20 / $0.66 | 187 | 100.0% |
| 29 | StreamLakefp8# | 3.36% | 19k | $0.12 / $0.48 | 120 | 100.0% |
| 30 | Open Inferencefp4# | 3.81% | 96k | $0.02 / $0.13 | 44 | 99.6% |
| 31 | Mistralfp8# | 10.67% | 2.0k | $0.30 / $1.20 | 198 | 99.4% |
| 32 | Phala# | 13.29% | 7.2k | $0.20 / $1.20 | 103 | 99.4% |
| 33 | SiliconFlowfp8# | 20.67% | 15k | $0.30 / $1.20 | 147 | 99.8% |
Recommended OpenRouter settings for DeepSeek V4.1 Flash 1017
{
"provider": {
"order": [
"reka",
"deepinfra/fp8/flex",
"inference-net/fp8",
"inference-net/us"
],
"ignore": [
"cohere",
"streamlake/fp8",
"open-inference/fp4",
"mistral/fp8",
"phala",
"siliconflow/fp8"
]
}
}Give llms.txt to your agent
Get updates when its successor drops
And which provider to use.
For model makers
Broken structured outputs carry your model's name
Your users don't see the provider. They see your model failing.
as often on SiliconFlow (fp8) as on Reka AI, its best provider.
broken structured outputs a day across its 30 providers.
Releasing a model? Make your own API the one every provider is measured against.
Set the bar for your model↵