Which model for Laptop 8 Go RAM?
6 GB usable for the model (80% of RAM, CPU execution). 34 open-weight models fit in memory.
The margin goes to context: count ~1 GB per 8k tokens for a 7B model. A small margin means a short context.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Gemma 3n E4BGoogle · 4B | 1318 | Q4_K_M | 5 Go | +1 Go | slow (CPU) | ollama run gemma3n:e4b |
| Gemma 3 4BGoogle · 4B | 1303 | Q4_K_M | 5 Go | +1 Go | slow (CPU) | ollama run gemma3:4b |
| Llama 3.2 3BMeta · 3.21B | 1167 | Q4_K_MQ8_0 also fits · 6 Go | 4 Go | +2 Go | slow (CPU) | ollama run llama3.2:3b |
| Llama 3.2 1BMeta · 1.24B | 1111 | Q4_K_MFP16 also fits · 5 Go | 3 Go | +3 Go | slow (CPU) | ollama run llama3.2:1b |
Not yet ranked by the Arena
No Arena Elo linked: recent release, model absent from the Arena, or a match we have not made yet.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Qwen3 0.6BAlibaba · 0.6B | — | Q4_K_MFP16 also fits · 3 Go | 2 Go | +4 Go | slow (CPU) | ollama run qwen3:0.6b |
| Gemma 3 1BGoogle · 1B | — | Q4_K_MFP16 also fits · 4 Go | 3 Go | +3 Go | slow (CPU) | ollama run gemma3:1b |
| Phi-4 MiniMicrosoft · 3.8B | — | Q4_K_M | 4 Go | +2 Go | slow (CPU) | ollama run phi4-mini |
| Gemma 4 E4BGoogle · 4B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| MiniCPM5-2BOpenBMB · 2.5B | — | Q8_0 | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| Qwen3 4BAlibaba · 4.02B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | ollama run qwen3:4b |
| Gemma 4 E2BGoogle · 2B | — | FP16 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Granite 4.1 3BIBM · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Lucie 7BOpenLLM France · 6.7B | — | Q4_K_M | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Ministral 3 3BMistral AI · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| OLMo 3 7BAllen AI · 7B | — | Q4_K_M | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| SmolLM3 3BHugging Face · 3.1B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
Image, video, audio and 3D
These models do not generate text, so the text Arena does not rank them. Indicative order, not a quality ranking.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Kokoro-82MHexgrad · 0.08B | — | FP16 | 2 Go | +4 Go | slow (CPU) | Hugging Face ↗ |
| ChatterboxResemble AI · 0.5B | — | FP16 | 3 Go | +3 Go | slow (CPU) | Hugging Face ↗ |
| Parakeet TDT 0.6B v3NVIDIA · 0.6B | — | FP16 | 3 Go | +3 Go | slow (CPU) | Hugging Face ↗ |
| TripoSRStability AI · 0.5B | — | FP16 | 3 Go | +3 Go | slow (CPU) | Hugging Face ↗ |
| Stable Fast 3DStability AI · 1B | — | FP16 | 4 Go | +2 Go | slow (CPU) | Hugging Face ↗ |
| CogVideoX 5BZhipu AI · 5B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| FLUX.2 [klein] 4BBlack Forest Labs · 4B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| TRELLISMicrosoft · 1.2B | — | FP16 | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| TRELLIS.2Microsoft · 4B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| Wan 2.2 5BAlibaba · 5B | — | Q4_K_M | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| Whisper large-v3OpenAI · 1.55B | — | FP16 | 5 Go | +1 Go | slow (CPU) | Hugging Face ↗ |
| Dia 1.6BNari Labs · 1.6B | — | FP16 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Hunyuan3D 2.1Tencent · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Ming-Image 0.1 DesignAnt Group · 6B | — | Q4_K_M | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Orpheus 3BCanopy Labs · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Voxtral Mini 3BMistral AI · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| YuE2-3Bm-a-p (Multimodal Art Projection) · 3B | — | Q8_0 | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
| Z-Image TurboAlibaba · 6B | — | Q4_K_M | 6 Go | +0 Go | slow (CPU) | Hugging Face ↗ |
An “ollama run” command without a suffix pulls the Q4_K_M build from the Ollama library: the row then shows that quantization and its memory, and flags whether FP16 or Q8_0 would also fit. Exception: a model published in a native format (MXFP4 for gpt-oss, FP8 for DeepSeek V4.1) exists in one build only, which is what the command loads. via Hugging Face or an explicit Ollama tag.
Share this verdict
One link, the same shortlist, the same reasons.