of DeepSeek V4.1 Flash 1017's structured outputs fail on Phala.
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 ↓
| # | Model | Broken structured outputs | Rank | Best for this model | Structured outputs a day |
|---|---|---|---|---|---|
| 1 | Qwen3.5 397B A17B | 0.83% | 2 of 7 | AtlasCloud (fp8) 0.65% | 5.3k |
| 2 | Qwen3.8 27B | 0.86% | 4 of 16 | NEAR AI (fp8) 0.25% | 26k |
| 3 | GLM 5.3 Flashnvfp4 | 1.38% | 18 of 29 | Parasail (fp4) 0.28% | 31k |
| 4 | DeepSeek V3.2 | 1.39% | 3 of 5 | DeepInfra (fp4) 0.67% | 6.3k |
| 5 | gpt-oss-120b | 2.47% | 14 of 18 | MARA 0.76% | 8.7k |
| 6 | Qwen3.6 35B A3B | 2.90% | 7 of 8 | Parasail (fp8) 0.09% | 13k |
| 7 | Kimi K3 | 3.14% | 16 of 20 | Parasail (fp4) 0.20% | 5.5k |
| 8 | DeepSeek V4 Flash 0731 | 6.14% | 17 of 18 | AtlasCloud (fp4) 0.16% | 16k |
| 9 | DeepSeek V4.1 Flash 1017 | 13.29% | 32 of 33 | Reka AI 0.40% | 7.2k |
Get updates when a new model drops
And which provider to use.
For inference providers
Structured outputs don't have to break
The same provider can handle structured outputs well on one model and struggle on the next. Implementations diverge on strings, optionals, composition and conditionals (the April announcement): the schema one provider enforces, another ignores.
on Qwen3.5 397B A17B, its best model.
models, where the settings tell OpenRouter to skip Phala.
dotjson plugs into your inference stack, so every model you serve gets perfect structured outputs from day 0.
Fix Phala's structured outputs↵