Machine

RTX 4070 · Ryzen 9 7900X #0b901106

283 runs of 47 models since Sep 18, 2026. Last run Nov 2, 2026.

ShareRedditX
283
runs
47
models
177
configs
49
configs run 2+ times

Hardware

GPU
NVIDIA GeForce RTX 4070 12 GB
CPU
AMD Ryzen 9 7900X 12-Core Processor 12 cores
RAM
63 GB
OS
Windows 11 Education 26H2
Tools
llama.cpp 142 · Unsloth Studio 131 · OpenAI-compatible 8 · Strata 2
Client
0.4.98
All NVIDIA GeForce RTX 4070 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 read0204060801002050100200500tok/s (log scale) →quality →outliers sit at the edges

Medians of each model's best config, 45 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.

swift-iq2_xs
2 runs · 1 config
★ bestStrata32k ctxthinking on
×2 runs
87.987.3–88.5
steady
91.5 tok/s
90.1–92.9
peculiar-ragdoll/Cyber-Tiel-Coder-35B-A3B-GGUF-MTP
5 runs · 4 configs
★ bestUD-IQ4_XSUnsloth Studio64k ctxthinking oneffort mediumKV q4_0
×2 runs
85.483.9–86.9
steady
78.9 tok/s
76.3–81.5
Ternary-Bonsai-2-27B-PTQ1_0
5 runs · 3 configs
★ bestQ1_0OpenAI-compatible64k ctxthinking default
×2 runs
84.279.9–88.6
varies
55.5 tok/s
55.3–55.8
ukisai/Swift-Bonsai-2-GGUF
3 runs · 2 configs
★ bestTQ1_0llama.cpp128k ctxthinking oneffort medium
×2 runs
84.182.1–86.1
steady
65.2 tok/s
65.2–65.3
prism-ml/Ternary-Bonsai-2-27B-gguf
15 runs · 12 configs
★ bestPTQ1_0llama.cpp64k ctxthinking default
×2 runs
83.080.0–85.9
steady
60.7 tok/s
60.0–61.5
ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF
6 runs · 4 configs
★ bestllama.cpp32k ctxthinking oneffort xhigh
×3 runs
81.976.9–87.7
varies
38.8 tok/s
38.5–39.8
protoLabsAI/Ornith-1.5-9B-MTP-GGUF
4 runs · 4 configs
★ bestIQ4_XSUnsloth Studio64k ctxthinking onKV f16
78.1
1 rated run
92.5 tok/s
unsloth/Ornith-1.0-9B-GGUF
5 runs · 4 configs
★ bestUD-Q4_K_XLUnsloth Studio32k ctxthinking defaultKV f16
×2 runs
75.774.2–77.2
steady
63.3 tok/s
59.1–67.4
unsloth/gemma-4-12B-it-qat-GGUF
5 runs · 5 configs
★ bestUD-Q4_K_XLUnsloth Studio100k ctxthinking onKV f16
73.8
1 rated run
50.4 tok/s
mistralai/Ministral-3-8B-Instruct-2512-GGUF
4 runs · 3 configs
★ bestQ4_K_Mllama.cpp64k ctxthinking default
×2 runs
72.871.8–73.8
steady
66.0 tok/s
63.9–68.2

Compare with another machine

Run history

Show all 283 runs ▾
Sep 18, 2026
19:30LFM2.5-1.2B-Instruct-GGUF336 tok/s30.119:28LFM2.5-1.2B-Thinking-GGUF314 tok/s30.219:26LFM2.5-1.2B-Thinking-GGUF319 tok/s31.219:23LFM2.5-VL-1.6B-GGUF×21053 tok/s18.417:04unknown41.0 tok/s66.016:35Qwen3.8-27B-GGUF×439.6 tok/s53.514:18NVIDIA-Nemotron-3-Nano-4B-GGUF95.0 tok/s51.314:15NVIDIA-Nemotron-3-Nano-4B-GGUF91.9 tok/s41.614:13NVIDIA-Nemotron-3-Nano-4B-GGUF91.6 tok/s39.514:08Qwen3.5-0.8B-MTP-GGUF237 tok/s26.514:06Qwen3.5-0.8B-MTP-GGUF241 tok/s22.114:02Qwen3.5-9B-MTP-GGUF×299.3 tok/s69.013:46LFM2.5-1.2B-Thinking-GGUF309 tok/s29.113:44LFM2.5-1.2B-Thinking-GGUF×2331 tok/s33.513:37gemma-4-12B-it-qat-GGUF50.1 tok/s69.713:29gemma-4-12B-it-qat-GGUF51.4 tok/s67.813:23LFM2.5-1.2B-Instruct-GGUF337 tok/s32.713:13Ornith-1.0-9B-GGUF60.1 tok/s72.712:59LFM2.5-8B-A1B-GGUF240 tok/s44.812:45Qwen3.8-27B-GGUF42.1 tok/s76.012:26gemma-4-E4B-it-qat-mobile-GGUF72.6 tok/s23.412:19gemma-4-E4B-it-qat-mobile-GGUF73.6 tok/s29.512:10LFM2.5-1.2B-Thinking-GGUF304 tok/s27.812:08LFM2.5-1.2B-Thinking-GGUF330 tok/s32.312:05LFM2.5-1.2B-Instruct-GGUF×2382 tok/s36.912:00LFM2.5-8B-A1B-GGUF241 tok/s52.611:46LFM2.5-8B-A1B-Q4_K_M276 tok/s50.911:42LFM2.5-8B-A1B-Q4_K_M×2284 tok/s47.811:25gpt-oss-20b-UD-Q4_K_XL117 tok/s41.711:16LFM2.5-8B-A1B-Q4_K_M282 tok/s55.0