Model cards¶
One card per model: what its numbers mean for a game, and where it sits among the other measured models.
- Bonsai 8B (Q1_0 1-bit) · general guard 31.6% · world guard 60.5% · generation 238 tok/s · VRAM 1.90 GB
- Bonsai 8B (Q2_0 2-bit) · general guard 62.8% · world guard 86.4% · generation 221 tok/s · VRAM 3.05 GB
- Gemma 4 E2B Instruct QAT (UD-Q4_K_XL) · general guard 75.0% · world guard 76.5% · generation 259 tok/s · VRAM 2.81 GB
- Gemma 4 E4B Instruct (Q4_K_M) · general guard 90.8% · world guard 90.1% · generation 143 tok/s · VRAM 5.78 GB
- Granite 4.2 3B Instruct (Q4_K_M) · general guard 26.5% · world guard 72.8% · generation 200 tok/s · VRAM 2.74 GB
- Granite 4.2 8B Instruct (Q4_K_M) · general guard 21.9% · world guard 87.7% · generation 112 tok/s · VRAM 6.35 GB
- Ling 3.0 Tiny (Q4_K_M) · general guard 34.2% · world guard 61.7% · generation 267 tok/s · VRAM 5.03 GB
- Llama 3.1 8B Instruct (Q4_K_M) · general guard 77.0% · world guard 80.2% · generation 134 tok/s · VRAM 5.60 GB
- Llama 3.2 3B Instruct (Q4_K_M) · general guard 32.1% · world guard 64.2% · generation 256 tok/s · VRAM 2.58 GB
- Ministral 3 3B Instruct (Q4_K_M) · general guard 41.8% · world guard 71.6% · generation 189 tok/s · VRAM 2.69 GB
- Ministral 3 8B Instruct (Q4_K_M) · general guard 59.7% · world guard 82.7% · generation 105 tok/s · VRAM 5.91 GB
- Qwen 3.5 4B (Q4_K_M) · general guard 78.6% · world guard 90.1% · generation 174 tok/s · VRAM 3.01 GB
- SmolLM2 1.7B Instruct (Q4_K_M) · general guard 35.7% · world guard 28.4% · generation 354 tok/s · VRAM 1.98 GB