of DeepSeek V4 Pro 0813's structured outputs fail on CoreWeave (fp8).
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.
| # | Provider | Structured outputs a day | $ in / out per M | Tokens/s | Uptime | |
|---|---|---|---|---|---|---|
| 1 | Cloudflare# | 0.39% | 7.0k | $1.32 / $3.96 | 36 | 99.7% |
| 2 | Nebius Token Factoryfp8# | 0.47% | 3.3k | $1.32 / $3.96 | 70 | 81.3% |
| 3 | Ionstream# | 0.53% | 20k | $0.37 / $2.93 | 36 | 100.0% |
| 4 | Together# | 0.83% | 9.6k | $1.32 / $3.96 | 119 | 99.4% |
| 5 | DeepInfrafp8# | 1.09% | 13k | $1.30 / $2.60 | 52 | 99.8% |
| 6 | DekaLLM# | 1.43% | 2.1k | $0.88 / $3.80 | 43 | 99.2% |
| 7 | AtlasCloudfp8# | 1.48% | 6.6k | $1.32 / $3.96 | 24 | 99.7% |
| 8 | Baidu Qianfanfp8# | 1.62% | 36k | $0.28 / $0.85 | 41 | 100.0% |
| 9 | Sail Researchfp4# | 1.81% | 62k | $0.68 / $1.20 | 49 | 100.0% |
| 10 | Wafermxfp4# | 2.02% | 46k | $0.12 / $4.00 | 96 | 99.7% |
| 11 | Alibaba Cloud Int.# | 3.60% | 9.3k | $0.58 / $1.74 | 26 | 99.0% |
| 12 | Parasailfp8# | 3.61% | 1.5k | $1.32 / $3.96 | 34 | 99.6% |
| 13 | CoreWeavefp8# | 4.69% | 13k | $1.31 / $3.96 | 128 | 99.1% |
Also ignored, under 1,000 structured outputs a day: DigitalOcean 3.29%, Venice 5.32%.
{
"provider": {
"order": [
"cloudflare",
"nebius/fp8",
"ionstream"
],
"ignore": [
"digitalocean",
"alibaba",
"parasail/fp8",
"coreweave/fp8",
"venice"
]
}
}Give llms.txt to your agent
And which provider to use.
Your users don't see the provider. They see your model failing.
of DeepSeek V4 Pro 0813's structured outputs fail on CoreWeave (fp8).
as often on CoreWeave (fp8) as on Cloudflare, its best provider.
broken structured outputs a day across its 13 providers.
Releasing a model? Make your own API the one every provider is measured against.
Set the bar for your model↵