of DeepSeek V4.1 Flash's structured outputs fail on Baseten.
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 | DeepInfrafp8/flex# | 0.35% | 254k | $0.11 / $0.34 | 108 | 98.0% |
| 2 | DeepInfrafp8/priority# | 0.40% | 60k | $0.21 / $0.63 | 113 | 53.8% |
| 3 | Modal# | 0.44% | 10k | $0.30 / $1.20 | 99 | 100.0% |
| 4 | DeepInfrafp8# | 0.52% | 2.2M | $0.14 / $0.42 | 104 | 100.0% |
| 5 | AtlasCloudfp8# | 0.56% | 879k | $0.14 / $0.54 | 99 | 99.8% |
| 6 | Venicefp8# | 0.72% | 176k | $0.30 / $1.20 | 177 | 99.9% |
| 7 | inference.netfp8# | 0.80% | 3.0M | $0.17 / $0.59 | 118 | 100.0% |
| 8 | Morphfp8# | 0.81% | 1.8M | $0.02 / $0.19 | 70 | 100.0% |
| 9 | Baidu Qianfanfp8# | 0.90% | 2.1M | $0.04 / $0.18 | 124 | 99.8% |
| 10 | Open Inferencefp4# | 0.94% | 687k | $0.02 / $0.14 | 109 | 99.6% |
| 11 | Nebius Token Factoryfp8# | 0.99% | 11k | $0.30 / $1.20 | 100 | 99.2% |
| 12 | Fireworks# | 1.19% | 30k | $0.30 / $1.20 | 278 | 99.5% |
| 13 | Sail Researchfp4# | 1.52% | 447k | $0.09 / $0.25 | — | 100.0% |
| 14 | DigitalOcean# | 1.62% | 887k | $0.23 / $0.90 | 198 | 100.0% |
| 15 | Fireworks Spotspot# | 1.92% | 704k | $0.21 / $0.84 | 43 | 100.0% |
| 16 | Together# | 1.93% | 1.5M | $0.30 / $1.20 | 250 | 100.0% |
| 17 | Relacefp4# | 2.03% | 927k | $0.01 / $0.53 | 105 | 99.7% |
| 18 | Makorafp8# | 2.17% | 43k | $0.26 / $1.17 | 191 | 99.9% |
| 19 | Reka AI# | 2.19% | 76k | $0.10 / $0.84 | 82 | 98.5% |
| 20 | Crusoefp8# | 2.44% | 82k | $0.29 / $1.20 | 146 | 99.3% |
| 21 | Decartfp4# | 2.53% | 325k | $0.09 / $0.18 | — | 97.0% |
| 22 | Parasailfp8# | 2.56% | 150k | $0.30 / $1.20 | 176 | 100.0% |
| 23 | Prime Intellect# | 2.68% | 90k | $0.30 / $1.20 | 172 | 100.0% |
| 24 | Wafermxfp4# | 2.93% | 1.2M | $0.04 / $0.50 | 220 | 100.0% |
| 25 | CoreWeavefp8# | 3.02% | 952k | $0.13 / $0.65 | 198 | 99.8% |
| 26 | Cohere# | 4.32% | 142k | $0.30 / $1.20 | 239 | 99.8% |
| 27 | Google Vertex# | 4.45% | 33k | $0.30 / $1.20 | 115 | 98.9% |
| 28 | Baseten# | 4.84% | 169k | $0.30 / $1.20 | 185 | 99.8% |
Also ignored, under 1,000 structured outputs a day: Mancer (fp8) 3.51%.
Recommended OpenRouter settings for DeepSeek V4.1 Flash
{
"provider": {
"order": [
"deepinfra/fp8/flex",
"deepinfra/fp8/priority",
"modal"
],
"ignore": [
"coreweave/fp8",
"mancer/fp8",
"cohere",
"google-vertex/global",
"baseten"
]
}
}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 Baseten as on DeepInfra (fp8/flex), its best provider.
broken structured outputs a day across its 25 providers.
Releasing a model? Make your own API the one every provider is measured against.
Set the bar for your model↵