| Model | Verdict | Decode tok/s | Max ctx | Size | Chat | Coding | |
|---|---|---|---|---|---|---|---|
| Qwen/Qwen3-30B-A3B-Instruct-2507 BF16 | Fits | 31.512.6–34.6 | 256K | 30.5B | p85 | – | |
| google/gemma-4-31B-it unsloth/gemma-4-31B-it-GGUF · Q8_0 | Fits | 7.005.60–8.70 | 256K | 31.3B | p63 | – | |
| Qwen/Qwen3.5-27B unsloth/Qwen3.5-27B-GGUF · Q8_0 | Fits | 9.307.90–10.7 | 256K | 27.8B | p60 | p46 | |
| Qwen/Qwen3.8-27B unsloth/Qwen3.8-27B-GGUF · Q8_0 | Fits | 9.207.80–10.6 | 256K | 27.8B | p59 | p33 | |
| Qwen/Qwen3.6-27B unsloth/Qwen3.6-27B-GGUF · Q8_0 | Fits | 9.307.90–10.7 | 256K | 27.8B | p51 | p54 | |
| Qwen/Qwen3.5-35B-A3B unsloth/Qwen3.5-35B-A3B-GGUF · Q8_0 | Fits | 44.135.3–55.1 | 256K | 36B | p49 | p35 | |
| google/gemma-4-26B-A4B-it BF16 | Fits | 27.911.2–30.7 | 256K | 25.8B | p49 | – | |
| Qwen/Qwen3.5-9B unsloth/Qwen3.5-9B-GGUF · Q8_0 | Fits | 28.624.3–32.9 | 256K | 9.7B | p46 | – | |
| ornith-ai/Ornith-1.5-9B AtomicChat/Ornith-1.5-9B-GGUF · Q8_0 | Fits | 28.624.3–32.9 | 256K | 9.7B | p46 | p24 | |
| Qwen/Qwen3.6-35B-A3B unsloth/Qwen3.6-35B-A3B-MTP-GGUF · Q8_0 | Fits | 43.334.7–54.2 | 256K | 36B | p43 | p41 | |
| openbmb/MiniCPM-V-4.7-35B-A3B openbmb/MiniCPM-V-4.7-35B-A3B-gguf · Q8_0 | Fits | 44.235.4–55.3 | 256K | 35.2B | p42 | – | |
| ornith-ai/Ornith-1.0-35B BF16 | Fits | 35.114.0–38.6 | 256K | 35.1B | – | p40 | |
| openai/gpt-oss-20b MXFP4 | Fits | 48.319.3–53.1 | 128K | 21.1B | p39 | p15 | |
| Qwen/Qwen3.5-4B unsloth/Qwen3.5-4B-GGUF · Q8_0 | Fits | 49.241.8–56.6 | 256K | 4.7B | p39 | – | |
| google/gemma-4-12B-it unsloth/gemma-4-12b-it-GGUF · Q8_0 | Fits | 17.314.7–19.9 | 256K | 12B | p38 | – | |
| Qwen/Qwen3-4B-Instruct-2507 BF16 | Fits | 25.610.3–28.2 | 256K | 4B | p38 | – | |
| nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 BF16 | Fits | 67.527.0–74.2 | 256K | 32.9B | p37 | p9 | |
| nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 MIXED_PRECISION | Fits | 86.634.7–95.3 | 1M | 47.1B | p36 | p9 | |
| nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 BF16 | Fits | 69.627.8–76.6 | 256K | 31.6B | p33 | – | |
| poolside/Laguna-XS-2.1 BF16 | Fits | 39.315.7–43.3 | 256K | 33.4B | – | p30 | |
| poolside/Laguna-XS.2 BF16 | Fits | 39.315.7–43.3 | 256K | 33.4B | – | p25 | |
| ornith-ai/Ornith-1.0-9B unsloth/Ornith-1.0-9B-GGUF · Q8_0 | Fits | 28.724.4–33.0 | 256K | 9.4B | – | p20 | |
| ornith-ai/Ornith-1.5-35B-A3B-NVFP4 MIXED_PRECISION | Fits | 10140.4–111 | 256K | 34B | – | – | |
| nvidia/Qwen3.6-35B-A3B-NVFP4 MIXED_PRECISION | Fits | 10140.4–111 | 256K | 34B | – | – | |
| Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 FP8 | Fits | 47.418.9–52.1 | 256K | 30.5B | – | – | |
| prism-ml/Bonsai-27B-mlx-1bit MLX-1BIT | Fits | 38.715.5–42.6 | 256K | 26.4B | – | – | |
| ornith-ai/Ornith-1.0-35B-GGUF Q8_0 | Fits | 37.530.0–46.9 | 256K | 34.7B | – | – | |
| ornith-ai/Ornith-1.5-35B-A3B-GGUF Q8_0 | Fits | 36.829.4–46.0 | 256K | 35.5B | – | – | |
| Qwen/Qwen3-30B-A3B BF16 | Fits | 31.512.6–34.6 | 40K | 30.5B | – | – | |
| Qwen/Qwen3-Coder-30B-A3B-Instruct BF16 | Fits | 31.512.6–34.6 | 256K | 30.5B | – | – | |
| ornith-ai/Ornith-1.5-9B-GGUF Q8_0 | Fits | 28.023.8–32.3 | 256K | 9.2B | – | – | |
| prism-ml/Ternary-Bonsai-27B-mlx-2bit MLX-2BIT | Fits | 26.810.7–29.5 | 256K | 26.4B | – | – | |
| Qwen/Qwen3-14B unsloth/Qwen3-14B-GGUF · Q8_0 | Fits | 13.711.0–17.1 | 40K | 14.8B | – | – | |
| Qwen/Qwen2.5-14B-Instruct Qwen/Qwen2.5-14B-Instruct-GGUF · Q8_0 | Fits | 13.210.6–16.5 | 32K | 14.8B | – | – | |
| RadixArk/Qwen3.8-27B-NVFP4 MIXED_PRECISION | Fits | 12.04.80–13.2 | 256K | 26.4B | – | – | |
| Qwen/Qwen2.5-Coder-14B-Instruct Qwen/Qwen2.5-Coder-14B-Instruct-GGUF · Q4_K_M | Fits | 11.99.50–14.8 | 32K | 14.8B | – | – | |
| farbodtavakkoli/OTel-2.0-LLM-31B-IT wfakhri/OTel-2.0-LLM-31B-IT-GGUF · Q8_0 | Fits | 7.005.60–8.70 | 256K | 31.3B | – | – | |
| Qwen/Qwen2.5-Coder-32B-Instruct Qwen/Qwen2.5-Coder-32B-Instruct-GGUF · Q3_K_M | Fits | 6.805.50–8.50 | 32K | 32.8B | – | – | |
| Qwen/Qwen2.5-32B-Instruct Qwen/Qwen2.5-32B-Instruct-GGUF · Q8_0 | Fits | 6.305.10–7.90 | 32K | 32.8B | – | – | |
| Qwen/Qwen3-32B BF16 | Fits | 4.001.60–4.40 | 40K | 32.8B | – | – |
Quality = benchmark percentile against all indexed models (p50 = median). Speeds are single-user decode ranges.