OLMo 2 32B
Allen AI · 32B · Dense
Fully open research model by Allen AI
3.7K
148
2025-03
4K tokens
Use cases
Your hardware
16 GB Reference Desktop
Quantization options
What is quantization?| Quant | Bits | VRAM | Disk | Quality | Status |
|---|---|---|---|---|---|
| Q2_K | 2 | 10.7 GB | 9.8 GB | low | Tight fit · ~19 t/s |
| Q3_K_M | 3 | 14.8 GB | 13.7 GB | moderate | Barely runs · ~14 t/s |
| Q4_K_M | 4 | 16.9 GB | 15.6 GB | good | Too heavy · ~8 t/s |
| Q5_K_M | 5 | 21 GB | 19.6 GB | good | Barely runs · ~4 t/s |
| Q6_K | 6 | 25.1 GB | 23.5 GB | excellent | Barely runs · ~3 t/s |
| Q8_0 | 8 | 33.3 GB | 31.3 GB | excellent | Barely runs · ~1 t/s |
| F16 | 16 | 66.1 GB | 62.6 GB | lossless | Too heavy · ~1 t/s |
About this model
OLMo 2 32B is a 32B dense transformer from Allen AI in the OLMo family, designed primarily for chat, reasoning workflows. Supports up to 4K context tokens; start at 4K–8K with Q8_0 KV cache on tight GPUs.
Hardware planning baselines start at 17.9 GB minimum memory and 29.8 GB recommended memory. At Q4_K_M, weights plus baseline runtime overhead use roughly 16.9 GB of RAM/VRAM and 15.6 GB of disk space. As a 32B dense checkpoint, every parameter participates in each token forward pass, meaning decoding speed scales linearly with GPU memory bandwidth and drops sharply if layers spill over PCIe into host DDR5 RAM.
Deploy OLMo 2 32B locally via llama.cpp with Flash Attention enabled. Under Apache License 2.0, review official weight cards before commercial redistribution.
Highlights
- Local Fit Baseline17.9 GB minimum and 29.8 GB recommended RAM/VRAM.
- Primary QuantQ4_K_M targets 16.9 GB memory and 15.6 GB disk.
- Context & Capabilities4K max context · Dense.
System requirements by device
Memory planning targets 17.9 GB minimum RAM/VRAM and 29.8 GB recommended memory. The baseline quant is Q4_K_M, estimated at 16.9 GB memory and 15.6 GB disk. Actual runtime usage shifts with context length, KV cache precision, and OS background overhead.
| Device | Notes |
|---|---|
| Mac (Apple silicon) | Use unified memory headroom (~75% physical RAM max) as the primary fit signal. Start with Q4_K_M and close heavy browser tabs first. |
| Windows / Linux GPU | Keep weights plus KV cache inside dedicated VRAM so Windows WDDM does not spill layers over PCIe into Shared GPU Memory. |
| Phones & Compact Edge Boards | Not practical on phones (17.9 GB min RAM exceeds Android LMK and iOS Jetsam limits). Pick a 2B–4B model on mobile. |
Start with a short session and leave memory for the operating system. A file fitting on disk does not guarantee that it will load in memory.
Estimated memory by quantization
These are rough estimates, not tested minimum requirements. Check the actual file and runtime documentation; context, offloading and operating-system overhead can change the memory needed.
| Quant Tier | Estimated memory | Disk Size | Quality |
|---|---|---|---|
| Q2_K | 10.7 GB | 9.8 GB | low |
| Q3_K_M | 14.8 GB | 13.7 GB | moderate |
| Q4_K_M | 16.9 GB | 15.6 GB | good |
| Q5_K_M | 21.0 GB | 19.6 GB | good |
| Q6_K | 25.1 GB | 23.5 GB | excellent |
| Q8_0 | 33.3 GB | 31.3 GB | excellent |
| F16 | 66.1 GB | 62.6 GB | lossless |
How to run OLMo 2 32B on your device
Start with a short prompt and the smallest quant that still gives the quality you need. A clean first run is more useful than loading a larger file with no memory headroom.
Run OLMo 2 32B on Mac
- Install LM Studio or llama.cpp on macOS Apple Silicon.
- Pull Q4_K_M (16.9 GB) first and confirm it stays below ~75% of your Mac's unified memory.
- Enable Flash Attention and cap initial context at 8K until memory pressure stays below 29.8 GB.
- Consult the Mac guide for Activity Monitor tuning and Metal offload tips.
Run OLMo 2 32B on Windows
- Install LM Studio for Windows for visual VRAM inspection, or Ollama when you want a localhost API endpoint.
- Load Q4_K_M (~16.9 GB) and verify Dedicated GPU Memory stays within VRAM without spilling into Shared GPU Memory.
- Reserve at least 17.9 GB memory minimum and target 29.8 GB for comfortable context growth.
- Enable Q8_0 KV cache or drop one quant tier if generation slows down over PCIe.
OLMo 2 32B vs alternatives
| Model | Params | Context | Min RAM | License |
|---|---|---|---|---|
| OLMo 2 32B | 32B | 4K | 17.9 GB | Apache 2.0 |
| Qwen 2.5 32B | 32B | 128K | 17.9 GB | Apache 2.0 |
| Qwen 2.5 Coder 32B | 32B | 128K | 17.9 GB | Apache 2.0 |
| Qwen 3 32B | 32B | 128K | 17.9 GB | Apache 2.0 |
| DeepSeek R1 Distill 32B | 32B | 64K | 17.9 GB | MIT |
| EXAONE 4.0 32B | 32B | 128K | 17.9 GB | EXAONE AI |
Use this table to choose the smallest nearby model that still matches the job. A related model with lower RAM can be the better local choice when the current page fits only at the edge of available memory.
Frequently asked questions
- What is OLMo 2 32B?
- OLMo 2 32B is a 32B dense transformer from Allen AI, supporting up to 4K tokens with 17.9 GB minimum RAM and 29.8 GB recommended RAM.
- How much VRAM or RAM does OLMo 2 32B require?
- At Q4_K_M, OLMo 2 32B occupies roughly 16.9 GB of active memory and 15.6 GB on disk. Add 2–6 GB of spare headroom for KV cache buffers, runtime scratch space, and OS apps (29.8 GB recommended).
- What is the best way to run OLMo 2 32B locally?
- Run OLMo 2 32B in llama.cpp starting with Q4_K_M. Keep initial context at 4K–8K tokens and confirm 100% GPU/Metal offload before scaling context length.
- Can OLMo 2 32B run on Android or iPhone?
- OLMo 2 32B requires 17.9 GB minimum memory, placing it outside practical 8 GB smartphone limits. Run it on a 16 GB+ Mac or desktop GPU instead, and pick a 2B–4B checkpoint for phones.
- How does OLMo 2 32B compare to Qwen 2.5 32B?
- Qwen 2.5 32B (32B, 17.9 GB min RAM) sits closest in size. Compare active parameters, context ceiling (4K vs 128K), and license terms (Apache 2.0) before switching.