The 9B sibling of mammoth-1 (released 2026-10-21, MIT, a full fine-tune of Qwen3.5-9B) for fast, cheap decisions: same single shared-prefix prefill returning calibrated probabilities for every option. Scores: best model under 15B on the Jev Decision Index 0.4 (62.65, tied #8 of 208) and 65.2 on a replica of the Artificial Analysis Decision Index, level with d1-ultra (65). Division of labor with mammoth-1: reach for the 27B when broader general knowledge matters, the 9B when latency and cost drive routing and guardrail duty. Self-hosting requires the official setup.sh with the OneJev runtime (a single CUDA GPU).
0
Quantized builds
39
Benchmark records
Quantized builds
Community and official quantized releases; "linked records" counts only benchmarks tied to that exact release.
No quantized builds cataloged yet.
Benchmark data
37 configurations
| Hardware | Framework | Quant | Decode | VRAM used | Samples | Evidence |
|---|---|---|---|---|---|---|
| NVIDIA RTX 4090 | llama.cpp | IQ1_S | 180.8 tok/s | 10.04 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 4090 | llama.cpp | IQ3_XS | 136.3 tok/s | 13.82 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 4090 | vLLM | IQ2_XS | 179.7 tok/s | 11.17 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 3090 | Ollama | Q4_K_S | 69.8 tok/s | 16.84 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 3090 | Strata | IQ1_S | 163.9 tok/s | 10.04 GB | 1 | L2 Cross-framework verified |
| AMD Radeon RX 7900 XTX | vLLM | Q3_K_M | 95 tok/s | 14.95 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5090 | llama.cpp | Q4_K_S | 219 tok/s | 16.84 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5090 | Ollama | Q5_K_S | 174.8 tok/s | 19.11 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5090 | Ollama | Q8_0 | 139 tok/s | 23.84 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 5080 | llama.cpp | IQ3_XS | 196.5 tok/s | 13.82 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 3060 12GB | vLLM | Q2_K | 70.1 tok/s | 11.4 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 4080 Super | Ollama | Q3_K_L | 145.8 tok/s | 15.2 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX 4080 Super | Strata | IQ3_XS | 244.7 tok/s | 13.82 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | llama.cpp | Q3_K_L | 32.7 tok/s | 15.71 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | vLLM | Q6_K | 22.6 tok/s | 21 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | vLLM | Q8_0 | 19.6 tok/s | 23.84 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | llama.cpp | Q5_K_S | 59.5 tok/s | 19.11 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | Ollama | FP16 | 30.8 tok/s | 37.06 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | Ollama | Q5_K_S | 53.6 tok/s | 19.11 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | llama.cpp | Q3_K_S | 117.4 tok/s | 14.38 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | llama.cpp | Q5_K_S | 83.3 tok/s | 19.11 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | Ollama | Q6_K | 62.3 tok/s | 21 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | Ollama | Q3_K_M | 303.5 tok/s | 14.95 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | Strata | IQ3_S | 322.4 tok/s | 14.57 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | vLLM | FP16 | 25.8 tok/s | 37.06 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | vLLM | IQ2_XS | 68.6 tok/s | 11.17 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | Ollama | Q5_K_M | 36.6 tok/s | 19.68 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | Ollama | IQ1_S | 275.8 tok/s | 10.04 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | Ollama | IQ4_XS | 194.7 tok/s | 16.09 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | Strata | IQ3_XS | 354.5 tok/s | 13.82 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | Strata | Q4_K_S | 293.1 tok/s | 16.84 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | llama.cpp | Q4_K_M | 92.5 tok/s | 17.6 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | llama.cpp | Q4_K_S | 88.6 tok/s | 16.84 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | Ollama | Q6_K | 67.8 tok/s | 21 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | IQ4_XS | 53.7 tok/s | 16.09 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | Q4_K_M | 50.5 tok/s | 17.6 GB | 2 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | Q4_K_M | 49.5 tok/s | 17.6 GB | 2 | L2 Cross-framework verified |
39 of these records are not linked to a specific quantized build; the full set is on the search page.
Community discussion
No discussions yet — start the first one.
Start a discussion