Small language models ranked by eff (Glint Tiny-ML protocol: BLiMP, ARC-Easy and WikiText-2 byte perplexity; the size bonus in eff is scaled for 1K–150M parameters). Overall is the same mean without the size bonus. “measured” = scored by Fabryka, “reported” = numbers from the model's authors.
| # | Model | Author | Params | eff | Overall | Trust |
|---|---|---|---|---|---|---|
| 1 | Slayer139 1.01 | Fabryka AI (Arkadiusz Słota) | 139.3M | 78.39 | 78.15 | measured |
| 2 | GPT-X2-125M | AxiomicLabs | 125.1M | 78.26 | 77.67 | measured |
| 3 | Slayer149 Balanced ARC-E fine-tune (149M) | Fabryka AI (Kacper Wikieł) | 149.3M | 77.92 | 77.90 | measured |
| 4 | Supra-50M-Base | SupraLabs | 51.8M | 77.62 | 74.30 | measured |
| 5 | JugnuLM-110M-R2plus | altslate | 109.7M | 77.12 | 76.12 | measured |
| 6 | Ivme-Conversate-v3-Base | IvmeLabs | 24.8M | 77.05 | 71.64 | measured |
| 7 | GoLLeM-v5-128M-final-arcmix-16x768 | Fabryka AI (Arkadiusz Słota) | 122.8M | 76.94 | 76.30 | measured |
| 8 | JugnuLM-53M | altslate | 53.5M | 76.62 | 73.44 | measured |
| 9 | Haidass-143M-v1 | DALabCommunity | 143.1M | 76.48 | 76.33 | measured |
| 10 | min-spark | MinimaLabs | 5.8M | 76.25 | 67.08 | measured |
Top 10 (English) from https://track.fabryka.ai/api/leaderboard/models, fetched 2026-10-09 10:35 UTC; JSON sha256 861926ae3da250a8. Snapshot: the live table is at https://track.fabryka.ai/models.