AI PROVIDERS
Model comparison
Every extraction runs the primary models, and the shadow models run in the background. Their drafts are compared with what reviewers save.
Primary: TypeSafe Jev (fallback Clef (Workers AI)) · Text: none · Shadow: Clef-flash (Workers AI), Clef (Workers AI), Clairic (Workers AI, preview), Gemma 4 26B (Workers AI)
| Provider | Kind | Runs | Errors | Mean time | Mean tokens | Agreement with reviewers |
|---|---|---|---|---|---|---|
| TypeSafe Jev | typed | 8 primary | 0/8 | 0.4 s | 5650 | Type 100% (1/1)Color 100% (1/1)Material 100% (1/1)Style — |
| Clef (Workers AI) | typed | 8 shadow | 0/8 | 0.8 s | 3499 | Type 50% (1/2)Color 50% (1/2)Material 50% (1/2)Style 0% (0/1) |
| Clef-flash (Workers AI) | typed | 8 shadow | 0/8 | 1.6 s | 3499 | Type 100% (1/1)Color 100% (1/1)Material 100% (1/1)Style — |
| Clairic (Workers AI, preview) | text | 1 primary, 7 shadow | 0/8 | 0.6 s | 3083 | Brand 0% (0/1)Model 0% (0/1) |
| Gemma 4 26B (Workers AI) | text | 2 primary, 6 shadow | 3/8 | 5.7 s | 378 | Brand 0% (0/1)Model — |
Agreement counts only items saved as reviewed, and fields where the model or the reviewer has a value. The primary model has an advantage, because reviewers start from its draft. To switch providers, change TYPED_PROVIDER, TEXT_PROVIDER or SHADOW_PROVIDERS in wrangler.jsonc, or use “Re-run” on an item.