Machine
RTX 4060 Laptop · Core i7-14700HX #2d038440
156 runs of 29 models since Oct 29, 2026. Last run Nov 19, 2026.
156
runs
29
models
31
configs
19
configs run 2+ times
Hardware
- GPU
- NVIDIA GeForce RTX 4060 Laptop GPU 8 GB
- CPU
- Intel(R) Core(TM) i7-14700HX 20 cores
- RAM
- 15 GB
- OS
- Windows 11 Home Single Language 26H2
- Tools
- llama.cpp 156
- Client
- 0.4.98
Models on this machine
One dot per model, at its best config. Up is better answers, right is faster. Time to finish counts every token the model wrote, thinking included, so a model that talks a lot ends up further left.
Best trade-off: quality 60+, and nothing else here is both faster and betterOther models · dot size = number of runs
Medians of each model's best config, 29 models.
Best config per model
sort
Runs with the same model, quant, tool, context, thinking and KV cache count as one config. Numbers are the median of its runs, and each dot is one run on the 0–100 quality scale. ★ marks the best config with at least 2 rated runs.
configrunsqualityeach runspeed
NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K-DFLASH-NVFP4
11 runs · 1 config
★ bestQ6_Kllama.cpp32k ctxthinking default
×11 runs
73.163.7–76.2
varies
66.6 tok/s
63.2–67.8
NVIDIA-Nemotron-3.5-Bolt-E4B-UD-Q4_K_XL
10 runs · 1 config
★ bestQ4_Kllama.cpp32k ctxthinking default
×10 runs
71.964.5–75.7
varies
46.6 tok/s
44.0–47.8
NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K-DFLASH-BF16
5 runs · 1 config
★ bestQ6_Kllama.cpp32k ctxthinking default
×5 runs
71.670.5–74.9
steady
61.0 tok/s
58.7–63.5
Qwen3.8-9B-Q4_K_M
5 runs · 1 config
★ bestQ4_K_Mllama.cpp32k ctxthinking default
×5 runs
71.470.6–77.7
steady
34.2 tok/s
33.8–35.7
NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K
10 runs · 1 config
★ bestQ6_Kllama.cpp32k ctxthinking default
×10 runs
71.064.7–76.7
varies
42.1 tok/s
41.0–43.0
Qwen3.5-4B-Q5_K_M
1 run · 1 config
NVIDIA-Nemotron-3.5-Bolt-E4B-UD-Q4_K_XL-DFLASH-NVFP4
10 runs · 1 config
★ bestQ4_Kllama.cpp32k ctxthinking default
×10 runs
68.265.5–76.3
varies
60.9 tok/s
57.6–65.6
gemma-4-E4B-it-Q4_0
11 runs · 1 config
★ bestQ4_0llama.cpp32k ctxthinking default
×11 runs
67.956.4–72.8
varies1 slow run
89.2 tok/s
54.1–99.9
Spark-X2.6-4B-Q8_0
24 runs · 3 configs
★ bestQ8_0llama.cpp32k ctxthinking defaulteffort high
×10 runs
67.638.6–71.8
1 outlier
37.7 tok/s
33.9–41.4
One run of the best config scored 38.6, the other 9 sit between 60.9 and 71.8. The median isn't thrown off by it.
Qwen3.8-4B-Q5_K_M
5 runs · 1 config
★ bestQ5_K_Mllama.cpp32k ctxthinking default
×5 runs
66.056.6–70.2
varies
52.4 tok/s
50.6–55.1
Compare with another machine
Run history
Nov 2, 2026
20:14LFM3-1.7B-BF16×2llama.cpp33.9 tok/s58.7Nov 1, 2026
07:07mythos-nano-Q8_0llama.cpp48.9 tok/s44.606:46Ministral-3-8B-Claude-Opus-Reasoning-Distilled.i1-Q4_K_Mllama.cpp35.0 tok/s56.106:36qwen3.5-4B-super-coder.Q4_0×3llama.cpp66.1 tok/s51.806:19Spark-X2.6-4B-Q8_0×2llama.cpp37.9 tok/s68.705:20Spark-X2.6-4B-Q8_0×4llama.cpp36.1 tok/s64.503:56Spark-X2.6-4B-Q8_0×8llama.cpp37.7 tok/s66.3Oct 31, 2026
23:50Spark-X2.6-4B-Q8_0×10llama.cpp42.6 tok/s63.103:46NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K-DFLASH-NVFP4llama.cpp63.2 tok/s68.503:39Mamba-Codestral-7B-v0.1-Q5_0llama.cpp35.5 tok/s12.603:19Qwen2.5-Coder-7B-Instruct-Q5_K_M×2llama.cpp37.5 tok/s42.303:08microsoft.Phi-4-mini-instruct.Q8_0llama.cpp49.1 tok/s34.702:46Ministral-3-8B-Instruct-2512-Q4_K_M×6llama.cpp35.0 tok/s63.401:24Mellum2-12B-A2.5B-Instruct-MXFP4_MOE×10llama.cpp87.2 tok/s65.600:09GRAFT-10B-A2B-Q4_K_Mllama.cpp22.6 tok/s35.6Oct 30, 2026
23:55theprint-12b-moe-3a-q4_k_m×10llama.cpp34.0 tok/s31.719:56NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K-DFLASH-BF16×5llama.cpp61.0 tok/s71.619:00allenai.OLMoE-1B-7B-0125-Instruct.Q5_K_M×10llama.cpp185 tok/s29.918:28Llama-3.1-Tulu-3-8B-Q4_K_M×10llama.cpp42.4 tok/s47.405:56NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K-DFLASH-NVFP4×10llama.cpp66.6 tok/s73.805:10NVIDIA-Nemotron-3.5-Bolt-E4B-Q6_K×10llama.cpp42.1 tok/s71.003:56NVIDIA-Nemotron-3.5-Bolt-E4B-UD-Q4_K_XL×5llama.cpp45.8 tok/s72.303:23NVIDIA-Nemotron-3.5-Bolt-E4B-UD-Q4_K_XL-DFLASH-NVFP4×10llama.cpp60.9 tok/s68.202:26NVIDIA-Nemotron-3.5-Bolt-E4B-UD-Q4_K_XL×5llama.cpp46.7 tok/s70.900:31Qwen3.8-9B-Q4_K_M×4llama.cpp34.2 tok/s72.0Show all 156 runs ▾Show fewer ▴
Oct 29, 2026
23:51Qwen3.8-4B-Q5_K_M×4llama.cpp52.2 tok/s65.923:10gemma-4-E4B-it-Q4_0×8llama.cpp90.5 tok/s67.922:06DeepSeek-R1-Distill-Qwen-7B-Q4_K_Mllama.cpp38.8 tok/s37.021:58gemma-4-E4B-it-Q4_0×3llama.cpp89.2 tok/s67.921:22Qwen3-8B-Q4_K_Mllama.cpp32.7 tok/s61.620:35qwen2.5-coder-7b-instruct-q4_k_mllama.cpp43.3 tok/s43.320:28Qwen3.5-9B-Q4_K_Mllama.cpp32.5 tok/s62.320:10Qwen3.5-4B-Q5_K_Mllama.cpp47.7 tok/s69.419:10Qwen3.8-9B-Q4_K_Mllama.cpp34.6 tok/s71.419:00Qwen3.8-4B-Q6_Kllama.cpp47.8 tok/s65.418:46Qwen3.8-4B-Q5_K_Mllama.cpp55.1 tok/s69.0