Emu3.5
Emu3.5 is an open-weights language model from Beijing Academy of Artificial Intelligence / BAAI, released October 30, 2025. It has 34B parameters (dense). The context window is 32K tokens.
We introduce Emu3.5, a large-scale multimodal world model that natively predicts the next state across vision and language. Emu3.5 is pre-trained end-to-end with a unified next-token prediction objective on a corpus of vision-language interleaved data containing over 10 trillion tokens, primarily derived from …
Check it against your own PC →
Can you run Emu3.5 locally?
The smallest common GPU it runs on is 16 GB of VRAM (with 32 GB of RAM): IQ3_XXS, split GPU / CPU, roughly 8 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 5.2 GB | 12.7 GB | ~2 |
| 8 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 7.2 GB | 10.7 GB | ~3 |
| 12 GB VRAM | Very slow | IQ3_XXS (noticeable loss) | 11.2 GB | 6.7 GB | ~4 |
| 16 GB VRAM | Runs, slowly | IQ3_XXS (noticeable loss) | 15.2 GB | 2.7 GB | ~8 |
| 24 GB VRAM | Runs on GPU | IQ4_XS (good) | 23.0 GB | – | ~30 |
| 32 GB VRAM | Runs on GPU | Q5_K_M (excellent) | 29.1 GB | – | ~42 |
Ready-made quantizations on Hugging Face: wikeeyang/Emu35-NF4
Benchmark results
Epoch AI has not published results for Emu3.5 yet. New models are usually evaluated within days to weeks.