of Kimi K2.5's structured outputs fail on Venice.
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 | GLM 5.2fp8 | 0.23% | 1 of 22 | — | 32k |
| 2 | GLM 5.3 Flash | 0.54% | 8 of 29 | Parasail (fp4) 0.28% | 1.8M |
| 3 | DeepSeek V4.1 Flashfp8 | 0.72% | 6 of 28 | DeepInfra (fp8/flex) 0.35% | 176k |
| 4 | Gemma 4 26B A4Bbf16 | 0.84% | 7 of 11 | Prime Intellect (bf16) 0.23% | 132k |
| 5 | Qwen3 235B A22B Instruct 2507fp8 | 0.86% | 1 of 3 | — | 78k |
| 6 | Qwen3.8 27Bfp8 | 1.14% | 6 of 16 | NEAR AI (fp8) 0.25% | 5.8k |
| 7 | DeepSeek V4 Pro 0423 | 1.24% | 2 of 8 | DeepInfra (fp8) 0.36% | 5.2k |
| 8 | Qwen3.5 397B A17B | 1.34% | 4 of 7 | AtlasCloud (fp8) 0.65% | 9.8k |
| 9 | DeepSeek V4.1 Flash 1017fp8 | 1.45% | 20 of 33 | Reka AI 0.40% | 2.7k |
| 10 | Nemotron 3.5 Super VLfp4 | 1.45% | 1 of 6 | — | 15k |
| 11 | Gemma 4 31Bfp4 | 1.79% | 11 of 13 | ModelRun [by Modular] (fp4) 0.21% | 172k |
| 12 | DeepSeek V4 Flash 0731 | 2.54% | 12 of 18 | AtlasCloud (fp4) 0.16% | 54k |
| 13 | Kimi K2.6int4 | 2.65% | 4 of 13 | Parasail (int4) 1.53% | 4.1k |
| 14 | MiMo-V2.6-Flashfp8 | 3.31% | 7 of 8 | GMICloud (bf16) 0.21% | 1.3k |
| 15 | DeepSeek V4 Flash 0423 | 3.56% | 10 of 13 | Alibaba Cloud Int. (fp8) 0.66% | 334k |
| 16 | Kimi K2.5 | 8.64% | 3 of 3 | AtlasCloud (int4) 2.44% | 17k |
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 GLM 5.2, its best model.
models, where the settings tell OpenRouter to skip Venice.
dotjson plugs into your inference stack, so every model you serve gets perfect structured outputs from day 0.
Fix Venice's structured outputs↵