of Llama 3.3 70B Instruct's structured outputs fail on DeepInfra (fp8/turbo).
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 | SambaNova Turbo# | 0.16% | 16k | $0.45 / $0.90 | 105 | 98.0% |
| 2 | Parasailfp8# | 0.54% | 101k | $0.22 / $0.50 | 47 | 100.0% |
| 3 | AkashMLfp8# | 0.67% | 185k | $0.20 / $0.52 | 46 | 100.0% |
| 4 | CoreWeavefp16# | 0.73% | 31k | $0.71 / $0.71 | 85 | 100.0% |
| 5 | DeepInfrafp8/turbo# | 3.48% | 248k | $0.10 / $0.32 | 20 | 99.6% |
Recommended OpenRouter settings for Llama 3.3 70B Instruct
{
"provider": {
"order": [
"sambanova-turbo",
"parasail/fp8",
"akashml/fp8"
],
"ignore": [
"deepinfra/turbo"
]
}
}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 DeepInfra (fp8/turbo) as on SambaNova Turbo, its best provider.
broken structured outputs a day across its 5 providers.
Releasing a model? Make your own API the one every provider is measured against.
Set the bar for your model↵