Leaderboard
Real benchmarks from real hardware. Compare local LLM performance across different setups. Run the command below to contribute yours.
$
npm install -g metrillm@latest$
metrillmRequires Node 20+ and Ollama or LM Studio running
Or run without installing: npx metrillm@latest
Trending
Last 7 days- 1 qwen3.5-4b-super-coder 1
- 2 nvidia-nemotron-3-nano-4b 1
- 3 parable-qwen3-4b-claude-fable-5 1
- 4 Flint-UD-Q4_K_XL-GGUF:latest 1
- 5 Sage-A1.5-Q4_K_M-GGUF:latest 1
- 1 Gigabyte Technology Co., Ltd. H310M S2 2.0 3
- 2 Micro-Star International Co., Ltd. MS-7E51 2
- 1 Intel Core™ i3-9100F 3
- 2 AMD Ryzen 7 9700X 8-Core Processor 2
Benchmarks
500
total runs
Models
283
unique models
Families
44
model families
Hardware
50
unique CPUs
Filters
500 results
| # | CPU | Model | Size | RAM | Runtime | tok/s | TTFT | HW Fit | Quality | Global | Flags | Verdict | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Apple M2 Max | qwen3-30b-a3b-thinking-2507-claude-4.5-sonnet-high-reasoning-distill-mlx | 30B/MXFP4 | 96 GB | LM-STUDIOMLX | 79.6 | 1.7 s | 95 | 98 | 97 | THINK | Excellent | ||
| 2 | Apple M4 Max | qwen3-30b-a3b-thinking-2507-claude-4.5-sonnet-high-reasoning-distill-mlx | 30B/MXFP4 | 128 GB | LM-STUDIOMLX | 112.8 | 1.5 s | 95 | 96 | 96 | THINK | Excellent | ||
| 3 | AMD RYZEN AI MAX+ 395 | nemotron3:33b | 33.0B/Q4_K_M | 125 GB | OLLAMAGGUF | 62.7 | 1.0 s | 95 | 97 | 96 | ECOTHINK | Excellent | ||
| 4 | Apple M4 Pro | gpt-oss-20b | — | 64 GB | LM-STUDIOGGUF | 118.2 | 2.2 s | 94 | 95 | 95 | THINK | Excellent | ||
| 5 | Apple M2 Max | qwen3-30b-a3b-thinking-2507-claude-4.5-sonnet-high-reasoning-distill-mlx | 30B/MXFP4 | 96 GB | LM-STUDIOMLX | 80.4 | 1.8 s | 94 | 96 | 95 | THINK | Excellent | ||
| 6 | Intel Core™ Ultra 9 285K | gpt-oss:20b | 20.9B/MXFP4 | 47 GB | OLLAMAGGUF | 239.9 | 616 ms | 98 | 94 | 95 | THINK | Excellent | ||
| 7 | Apple M4 | openai/gpt-oss-20b | 20B/MXFP4 | 24 GB | LM-STUDIOMLX | 29.3 | 1.8 s | 99 | 93 | 95 | THINK | Excellent | ||
| 8 | AMD Ryzen 5 5500 | nemotron-3-nano:4b | 4.0B/Q4_K_M | 8 GB | OLLAMAGGUF | 107.6 | 1.1 s | 99 | 93 | 95 | THINK | Excellent | ||
| 9 | Cortex-X925 | gemma4:26b | 25.8B/Q4_K_M | 122 GB | OLLAMAGGUF | 67.4 | 529 ms | 100 | 92 | 94 | Excellent | |||
| 10 | Intel Xeon® E5-2637 v3 | qwen3-30b-a3b-t16-c32k-instruct:latest | 30.5B/Q4_K_M | 96 GB | OLLAMAGGUF | 116.0 | 575 ms | 100 | 91 | 94 | Excellent | |||
| 11 | Intel Core™ i7-14700F | GPT-20B:20b | 20.9B/MXFP4 | 64 GB | OLLAMAGGUF | 181.0 | 1.3 s | 96 | 93 | 94 | THINK | Excellent | ||
| 12 | Apple M4 Pro | gpt-oss:20b | 20.9B/MXFP4 | 48 GB | OLLAMAGGUF | 54.5 | 2.9 s | 92 | 95 | 94 | THINK | Excellent | ||
| 13 | Apple M4 | openai/gpt-oss-20b | 20B/MXFP4 | 32 GB | LM-STUDIOMLX | 40.7 | 1.3 s | 100 | 92 | 94 | THINK | Excellent | ||
| 14 | Apple M4 | qwen/qwen3-30b-a3b-2507 | 30B/4bit | 32 GB | LM-STUDIOMLX | 44.8 | 405 ms | 100 | 91 | 94 | Excellent | |||
| 15 | AMD RYZEN AI MAX+ 395 | gpt-oss:20b | 20.9B/MXFP4 | 64 GB | OLLAMAGGUF | 48.0 | 2.3 s | 91 | 95 | 94 | THINK | Excellent | ||
| 16 | Intel Core™ i5-14400F | gpt-oss:20b | 20.9B/MXFP4 | 32 GB | OLLAMAGGUF | 80.4 | 1.6 s | 97 | 93 | 94 | THINK | Excellent | ||
| 17 | Cortex-X925 | gpt-oss:20b | 20.9B/MXFP4 | 122 GB | OLLAMAGGUF | 60.5 | 3.1 s | 87 | 95 | 93 | THINK | Excellent | ||
| 18 | Apple M4 Pro | nemotron-3-nano | — | 64 GB | LM-STUDIOGGUF | 93.3 | 257 ms | 100 | 90 | 93 | THINK | Excellent | ||
| 19 | AMD Ryzen 7 9700X 8-Core Processor | google/gemma-4-26b-a4b-qat | 26B/Q4_0 | 31 GB | LM-STUDIOGGUF | 109.7 | 156 ms | 96 | 92 | 93 | Excellent | |||
| 20 | Intel Core™ i7-14700F | GPT-OSS-20B:20b | 20.9B/MXFP4 | 128 GB | OLLAMAGGUF | 178.6 | 1.4 s | 92 | 93 | 93 | THINK | Excellent |
...
Don't miss new benchmarks
Get notified when new models and hardware configurations are tested. No spam, unsubscribe anytime.