Aion 1.0 Instruct 2.3B VRAM requirements
Microsoft · 2.29B params · MIT. Estimates, not vendor numbers.
Aion 1.0 Instruct 2.3B (2.29B parameters) needs about 2.9 GB of VRAM or unified memory at Q4, 3.1 GB at Q5 and 3.9 GB at Q8 (weights plus about 1.5 GB headroom, our estimate). The smallest common setup that runs it is 8GB GPU (RTX 4060).
| Quant | Download | Memory needed |
|---|---|---|
| Q4 | 1.4 GB | 2.9 GB |
| Q5 | 1.6 GB | 3.1 GB |
| Q8 | 2.4 GB | 3.9 GB |
Will it run on my machine?
| Setup | Runs? | Speed est. |
|---|---|---|
| 8GB GPU (RTX 4060) | Yes, Q8 | ~67 tok/s |
| 12GB GPU (RTX 3060 / 4070) | Yes, Q8 | ~124 tok/s |
| 16GB GPU (RTX 4060 Ti 16GB) | Yes, Q8 | ~71 tok/s |
| 24GB GPU (RTX 4090) | Yes, Q8 | ~249 tok/s |
| 32GB GPU (RTX 5090) | Yes, Q8 | ~442 tok/s |
| Mac 16GB (M4) | Yes, Q8 | ~30 tok/s |
| Mac 48GB (M4 Pro) | Yes, Q8 | ~67 tok/s |
| Mac 128GB (M4 Max) | Yes, Q8 | ~135 tok/s |
| PC, 32GB RAM, no GPU | Yes, Q8 | ~15 tok/s |
Best quant that fits each setup, with estimated generation speed from memory bandwidth. Real speed depends on the runtime, context length and settings.
Microsoft's new on-device small model, 32K context, built for summarizing, rewriting and chat. Open weights posted Nov 18, 2026; on Nov 19 it replaced Phi-4 mini as Microsoft Edge's built-in model. Microsoft says it isn't trained for world knowledge or complex multi-step reasoning.
FAQ
How much VRAM does Aion 1.0 Instruct 2.3B need?
About 2.9 GB at Q4, 3.1 GB at Q5 and 3.9 GB at Q8, including about 1.5 GB headroom. These are estimates from the parameter count on the official model card.
Can Aion 1.0 Instruct 2.3B run on 8GB of VRAM?
Yes, at Q8.
Can Aion 1.0 Instruct 2.3B run on a Mac?
Yes. Mac 16GB (M4), Mac 48GB (M4 Pro), Mac 128GB (M4 Max) can run it.
Which quant of Aion 1.0 Instruct 2.3B should I use?
Use the highest quant that fits with room to spare: Q8 is near full quality, Q5 is a good balance, and Q4 is the smallest that usually keeps quality acceptable.
Similar-size models
LFM3 1.7B (1.73B) · Ministral 3 3B (3B) · Phi-4 mini 3.8B (3.8B) · Qwen3 4B (4B) · Gemma 4 E4B (8B) · Ministral 3 8B (8B)