Compare LLM Models
Select 2-5 models to compare side by side.
Select models to compare
0 / 5 selected
Top downloads
Select 2-5 models
| Attribute |
🤖
Qwen3-4B
Qwen
|
|---|---|
| Company | Qwen |
| Release Date | 2025-04-29 |
| Parameters (B) | 4 |
| Architecture | Transformer |
| Context Window | 32768 |
| Input Modalities | text |
| Open Source | Yes |
| License | apache-2.0 |
| Score | 0 |
| VRAM (GB) | 8 |
| Compute | NVIDIA GPU with ~8GB VRAM for BF16/FP16 inference; 4-bit quantization reduces requirement to ~2–4GB; deployable via vLLM, SGLang, llama.cpp, or Ollama on consumer hardware |
| Benchmark: AIME-2024 | 73.8 |
| Benchmark: AIME-2025 | 65.6 |
| Benchmark: BBH | 72.6 |
| Benchmark: C-Eval | 77.5 |
| Benchmark: GPQA | 36.9 |
| Benchmark: GPQA-Diamond | 55.9 |
| Benchmark: GSM8K | 87.8 |
| Benchmark: IFEval | 81.9 |
| Benchmark: LiveCodeBench | 54.2 |
| Benchmark: MATH | 54.1 |
| Benchmark: MATH-500 | 97 |
| Benchmark: MBPP | 67 |
| Benchmark: MMLU | 73.0 |
| Benchmark: MMLU-Pro | 50.6 |
| Pricing (Input) | — |
| Pricing (Output) | — |