All models › microsoft
AesCode-32B
microsoft · on Arena since 29 Sept 2026
Download New · waiting for Arena votes
Arena score—Not voted on Arena yet
Size33.4 Bparameters
4-bit download20 GBweights only, Q4_K_M
Max context256Ktokens, from the model config
Where to use it for free
- Download and run it at home: official weights · GGUF files for Ollama and LM Studio
New: waiting for Arena votes
Released on 29 Sept 2026. Arena has not voted on this model yet. As soon as it does, the model moves into the ranking automatically, usually the next night. Until then you can still check its size, memory needs and speed below.
Speed on popular hardware
| Hardware | GPU memory | Fits? | Speed |
|---|---|---|---|
| RTX 3060 12 GB | 12 GB | GPU + 64 GB RAM | 3.7 tok/s |
| RTX 4060 Ti 16 GB | 16 GB | GPU + 64 GB RAM | 4.0 tok/s |
| RTX 3090 | 24 GB | GPU + 64 GB RAM | 9.4 tok/s |
| RTX 4090 | 24 GB | GPU + 64 GB RAM | 9.6 tok/s |
| RTX 5090 | 32 GB | Fits in GPU | 44 tok/s |
| RX 7900 XTX | 24 GB | GPU + 64 GB RAM | 9.5 tok/s |
| 2 × RTX 3090 | 48 GB | Fits in GPU | 22 tok/s |
| 4 × RTX 3090 | 96 GB | Fits in GPU | 21 tok/s |
| Mac M4 Max 128 GB | 96 GB | Fits in GPU | 13 tok/s |
| Ryzen AI Max+ 395 128 GB | 96 GB | Fits in GPU | 6.2 tok/s |
| 8 × H100 SXM | 640 GB | Fits in GPU | 67 tok/s |
Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 70% ÷ data read per token; real numbers vary by about ±30%.
GPU memory needed
| Level | 8K context | 32K context | 128K context | Quality vs full |
|---|---|---|---|---|
| Q3_K_M | 19 GB | 26 GB | 52 GB | +6.7% noticeable loss |
| Q4_K_M | 23 GB | 29 GB | 55 GB | +1.8% small loss |
| Q5_K_M | 26 GB | 33 GB | 59 GB | +0.6% near-lossless |
| Q6_K | 30 GB | 37 GB | 62 GB | +0.2% near-lossless |
| Q8_0 | 38 GB | 45 GB | 70 GB | ≈ 0% same as full |
| F16 | 70 GB | 76 GB | 102 GB | ≈ 0% same as full |
Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.