of GLM 5.3's structured outputs fail on DekaLLM.
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 | Together# | 0.25% | 53k | $1.40 / $4.40 | 94 | 100.0% |
| 2 | Basetenfp4# | 0.59% | 11k | $1.40 / $4.40 | 75 | 99.6% |
| 3 | Fireworks Spotspot# | 0.62% | 339k | $0.98 / $3.08 | — | 95.4% |
| 4 | Fireworks# | 0.73% | 14k | $1.40 / $4.40 | 79 | 99.5% |
| 5 | Modal# | 0.77% | 205k | $1.40 / $4.40 | 137 | 100.0% |
| 6 | Morphfp8# | 0.80% | 7.3k | $0.07 / $1.40 | 43 | 99.1% |
| 7 | io.netfp8# | 0.94% | 1.3k | $0.33 / $3.20 | 64 | 99.9% |
| 8 | Fireworksfast# | 1.11% | 27k | $2.10 / $6.60 | 152 | 99.0% |
| 9 | DeepInfrafp4# | 1.21% | 7.4k | $0.56 / $2.50 | 71 | 99.2% |
| 10 | AkashMLfp8# | 1.87% | 5.5k | $0.19 / $4.40 | 95 | 99.9% |
| 11 | Amazon Bedrock# | 1.91% | 1.5k | $1.68 / $5.28 | 147 | 82.0% |
| 12 | Waferfp4# | 2.19% | 271k | $0.07 / $1.40 | 108 | 99.8% |
| 13 | Makorafp4# | 2.74% | 7.4k | $0.14 / $3.00 | 81 | 98.0% |
| 14 | Mistralnvfp4/zdr# | 2.83% | 9.1k | $1.40 / $4.40 | 100 | 99.9% |
| 15 | Decartfp4# | 3.47% | 26k | $0.84 / $2.69 | 109 | 100.0% |
| 16 | Parasailfp8# | 4.00% | 7.1k | $1.40 / $4.40 | 101 | 99.7% |
| 17 | Sail Researchfp8# | 4.19% | 18k | $0.50 / $2.00 | 71 | 100.0% |
| 18 | Mistralnvfp4# | 4.34% | 3.8k | $1.40 / $4.40 | 136 | 100.0% |
| 19 | DigitalOcean# | 4.41% | 4.8k | $0.91 / $2.86 | 79 | 99.9% |
| 20 | Crusoefp4# | 4.97% | 18k | $1.40 / $4.40 | 116 | 100.0% |
| 21 | CoreWeavefp8# | 5.29% | 3.9k | $1.20 / $4.00 | 98 | 97.4% |
| 22 | Nebius Token Factoryfp4# | 6.30% | 1.1k | $1.40 / $4.40 | 98 | 97.6% |
| 23 | inference.netfp4# | 6.90% | 83k | $0.10 / $2.77 | 128 | 100.0% |
| 24 | Baidu Qianfanfp8# | 7.11% | 3.8k | $0.28 / $0.88 | 66 | 99.7% |
| 25 | Prime Intellect# | 10.28% | 9.0k | $0.98 / $3.08 | 96 | 99.8% |
| 26 | Reka AI# | 10.83% | 3.7k | $0.14 / $2.10 | 50 | 99.3% |
| 27 | Friendli# | 15.56% | 2.6k | $1.26 / $3.96 | 106 | 100.0% |
| 28 | DekaLLM# | 24.50% | 7.5k | $0.18 / $2.95 | 74 | 100.0% |
Also ignored, under 1,000 structured outputs a day: Cloudflare 4.76%, AtlasCloud (fp8) 11.24%, Venice 16.67%.
Recommended OpenRouter settings for GLM 5.3
{
"provider": {
"order": [
"together",
"baseten/fp4",
"fireworks/spot"
],
"ignore": [
"decart/fp4",
"parasail/fp8",
"sail-research/fp8",
"sail-research/us",
"mistral/nvfp4",
"digitalocean",
"cloudflare",
"crusoe/fp4",
"coreweave/fp8",
"nebius/fp4",
"inference-net/fp4",
"inference-net/us",
"baidu/fp8",
"primeintellect",
"reka",
"atlas-cloud/fp8",
"friendli",
"venice",
"dekallm"
]
}
}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 DekaLLM as on Together, 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↵