Machine

RTX 3060 · Ryzen 5 5600G #adfb9365

26 runs of 14 models since Oct 20, 2026. Last run Nov 11, 2026.

ShareRedditX
26
runs
14
models
21
configs
2
configs run 2+ times

Hardware

GPU
NVIDIA GeForce RTX 3060 12 GB
CPU
AMD Ryzen 5 5600G with Radeon Graphics 6 cores
RAM
30 GB
OS
Ubuntu 24.04.5 LTS
Tools
llama.cpp 26
Client
0.4.99
All NVIDIA GeForce RTX 3060 benchmarks

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
below 20 tok/s: slower than you read305070902050100tok/s (log scale) →quality →

Medians of each model's best config, 14 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.

Nex-N2.5-mini-APEX-I-MiniPlus-V2.1-GGUF
1 run · 1 config
★ bestllama.cpp16k ctxthinking default
84.5
1 rated run
53.3 tok/s
model
2 runs · 2 configs
★ bestllama.cpp16k ctxthinking default
83.9
1 rated run
53.7 tok/s
Spark-X2.6-4B-Precision-42pc
1 run · 1 config
★ bestllama.cpp193k ctxthinking defaulteffort high
76.5
1 rated run
56.0 tok/s
gemma-4-E4B-it-M-TS-Q4_K_M
5 runs · 2 configs
★ bestQ4_K_Mllama.cpp16k ctxthinking default
×4 runs
72.166.6–76.6
varies
80.5 tok/s
78.8–81.9
LFM3-1.7B-BF16
1 run · 1 config
★ bestllama.cpp256k ctxthinking default
68.5
1 rated run
83.5 tok/s
Spark-X2.6-4B-Q4_K_M
4 runs · 4 configs
★ bestQ4_K_Mllama.cpp16k ctxthinking oneffort high
67.3
1 rated run
68.1 tok/s
Ternary-Bonsai-2-27B-Abliterated-PTQ1_0
1 run · 1 config
★ bestQ1_0llama.cpp16k ctxthinking defaulteffort xhigh
65.0
1 rated run
41.8 tok/s
maritaca-ai_S5-nano-Q4_K_M
2 runs · 2 configs
★ bestQ4_K_Mllama.cpp32k ctxthinking on
62.8
1 rated run
72.3 tok/s
LFM3-1.7B-Heretic-Abliterated-Q6_K
1 run · 1 config
★ bestQ6_Kllama.cpp64k ctxthinking default
61.8
1 rated run
167 tok/s
S5-nano-Q8_0
1 run · 1 config
★ bestQ8_0llama.cpp8k ctxthinking off
54.5
1 rated run
57.3 tok/s

Compare with another machine

Run history