of GLM 5.3's structured outputs fail on Nebius Token Factory (fp4).
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 | MiniMax M3.1 Flashfp4 | 0.14% | 1 of 8 | — | 14k |
| 2 | GLM 5.1fp8 | 0.45% | 1 of 3 | — | 5.3k |
| 3 | DeepSeek V4 Pro 0813fp8 | 0.47% | 2 of 13 | Cloudflare 0.39% | 3.3k |
| 4 | Qwen3 30B A3B Instruct 2507fp8 | 0.65% | 1 of 3 | — | 231k |
| 5 | GLM 5.3 Flashfp4 | 0.98% | 13 of 29 | Parasail (fp4) 0.28% | 13k |
| 6 | DeepSeek V4.1 Flashfp8 | 0.99% | 11 of 28 | DeepInfra (fp8/flex) 0.35% | 11k |
| 7 | gpt-oss-120bfp4 | 1.05% | 3 of 18 | MARA 0.76% | 57k |
| 8 | Muse Spark Openfp4 | 1.31% | 2 of 4 | Together 0.68% | 1.1k |
| 9 | Beam 1.0 501B A23Bfp4 | 1.41% | 1 of 2 | — | 14k |
| 10 | DeepSeek V4.1 Flash 1017fp8 | 1.45% | 21 of 33 | Reka AI 0.40% | 32k |
| 11 | Gemma 3 27Bfp8 | 1.62% | 4 of 4 | NextBit (fp8) 0.74% | 180k |
| 12 | GLM 5.2fp4 | 5.44% | 19 of 22 | Venice (fp8) 0.23% | 6.4k |
| 13 | GLM 5.3fp4 | 6.30% | 22 of 28 | Together 0.25% | 1.1k |
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 MiniMax M3.1 Flash, its best model.
models, where the settings tell OpenRouter to skip Nebius Token Factory.
dotjson plugs into your inference stack, so every model you serve gets perfect structured outputs from day 0.
Fix Nebius Token Factory's structured outputs↵