of DeepSeek V4.1 Flash 1017's structured outputs fail on SiliconFlow (fp8).
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 | Qwen3.6 35B A3Bfp8 | 0.82% | 5 of 8 | Parasail (fp8) 0.09% | 1.5k |
| 2 | Qwen3 30B A3B Instruct 2507fp8 | 1.02% | 2 of 3 | Nebius Token Factory (fp8) 0.65% | 193k |
| 3 | DeepSeek V3.2fp8 | 1.54% | 4 of 5 | DeepInfra (fp4) 0.67% | 145k |
| 4 | Gemma 4 31Bfp8 | 1.57% | 10 of 13 | ModelRun [by Modular] (fp4) 0.21% | 60k |
| 5 | DeepSeek V4 Flash 0731fp8 | 1.71% | 9 of 18 | AtlasCloud (fp4) 0.16% | 91k |
| 6 | Gemma 4 26B A4Bfp8 | 2.05% | 9 of 11 | Prime Intellect (bf16) 0.23% | 74k |
| 7 | Kimi K2.5int4 | 3.19% | 2 of 3 | AtlasCloud (int4) 2.44% | 16k |
| 8 | Qwen3.5-9Bfp8 | 5.05% | 4 of 6 | Prime Intellect (bf16) 0.30% | 23k |
| 9 | gpt-oss-120bfp8 | 5.81% | 15 of 18 | MARA 0.76% | 6.7k |
| 10 | DeepSeek V3.1fp8 | 6.09% | 4 of 4 | MARA 0.12% | 19k |
| 11 | gpt-oss-20bfp8 | 9.76% | 6 of 8 | CoreWeave (fp4) 1.17% | 9.2k |
| 12 | DeepSeek V4.1 Flash 1017fp8 | 20.67% | 33 of 33 | Reka AI 0.40% | 15k |
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 Qwen3.6 35B A3B, its best model.
models, where the settings tell OpenRouter to skip SiliconFlow.
dotjson plugs into your inference stack, so every model you serve gets perfect structured outputs from day 0.
Fix SiliconFlow's structured outputs↵