Olmo 3 1125 32B — Hardware Requirements & GPU Compatibility
Chat
Olmo 3 1125 32B is a 32.2B-parameter open language model from Allen AI in the OLMo family. It supports a context window of up to 65,536 tokens. At BF16 it needs about 65.30 GB of VRAM — see which GPUs and Macs can run it below.
Olmo 3 1125 32B runs on a 96 GB+ Mac or a workstation GPU
Best local pick → Mac Studio (M3 Ultra, 96GB) · BF16 · 66.6 GB · ~9 tok/s
Macs from 96 GB (BF16 · ~9 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Olmo 3 1125 32B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve allenai/Olmo-3-1125-32B --max-model-len 16384 --tensor-parallel-size 8vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX PRO 6000 (96 GB) (29 GB headroom) · BF16
- Generation speed
- ~18 tok/s
- generation speed
- Cost per 1M output tokens
- $19.18
- per 1M output tokens
How Much VRAM Does Olmo 3 1125 32B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 65.3 GB | 67.7 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on RTX PRO 5000 Blackwell 72 GB |
Selected BF16
Brain floating point 1666.6GBat 16K context
- Weights
- 64.5 GB
- KV cache
- 1.9 GB
- Overhead
- 0.3 GB
Server-class: ~66.6 GB at 16K context · one RTX PRO 5000 Blackwell 72 GB
Which Hardware Can Run Olmo 3 1125 32B?
BF16 · 65.3 GBOlmo 3 1125 32B (BF16) requires 65.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 85+ GB is recommended. Using the full 66K context window can add up to 4.6 GB, bringing total usage to 69.9 GB. No consumer GPU has enough memory. 30 devices with unified memory can run Olmo 3 1125 32B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).
96 GB
fits · 29.4 GB free · up to 64K ctx
3 GPUs · 2 Mac classes
RTX PRO 6000 Blackwell Max-Q Workstation Edition~18Simulate output on RTX PRO 6000 Blackwell Max-Q Workstation EditionRTX PRO 6000 Blackwell Workstation Edition~18Simulate output on RTX PRO 6000 Blackwell Workstation EditionRTX PRO 6000 Blackwell Server Edition~16Simulate output on RTX PRO 6000 Blackwell Server EditionM3 Ultra*96 GB~9Simulate output on Mac Studio (M3 Ultra, 96GB)M3 Max*96 GB~3Simulate output on MacBook Pro 14" M3 Max (96 GB)128–144 GB
fits · 61.4 GB free · up to 64K ctx
4 GPUs · 2 Mac classes · 4 device families
GH200 Grace Hopper Superchip~49Simulate output on GH200 Grace Hopper SuperchipH200 NVL~48Simulate output on H200 NVLH200 SXM~48Simulate output on H200 SXMM2 Ultra*128 GB~9Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~7Simulate output on Mac Studio M5 Max (128 GB)GB10~3Simulate output on ASUS Ascent GX10Jetson T5000 (Thor)~3Simulate output on NVIDIA Jetson AGX Thor Developer Kit
By VRAM tier · typical GPUs and Macs at BF16
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 66.6 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Alternatives to Olmo 3 1125 32B
OLMo family · 20 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Hy Memory 30B A3BTencent | — | 19.1 GB | ~184 | same size |
| Llm Jp 4.1 32B A3b Thinkingllm-jp | — | 20.1 GB | ~188 | same size |
| Llm Jp 4.1 33B Thinkingllm-jp | — | 22.5 GB | ~35 | same size |
| Agnes 3.0 FlashAgnes-AI | — | 22.1 GB | ~35 | same size |
| NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA | — | 19.3 GB | ~197 | same size |
| Hy MT2 30B A3BTencent | — | 19.1 GB | ~170 | same size |
Fine-tunes & variants(1)
Specifications
- Publisher
- Allen AI
- Family
- OLMo
- Parameters
- 32.2B
- Architecture
- Olmo3ForCausalLM
- Context Length
- 65,536 tokens
- Vocabulary Size
- 100,278
- Release Date
- 2025-11-04
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Olmo 3 1125 32B need?
Olmo 3 1125 32B requires 65.3 GB of VRAM at BF16. Full 66K context adds up to 4.6 GB (69.9 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 32.2B × 16 bits ÷ 8 = 64.5 GB
KV Cache + Overhead ≈ 0.8 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 5.4 GB (at full 66K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Olmo 3 1125 32B?
No — Olmo 3 1125 32B requires at least 65.3 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
Can I run Olmo 3 1125 32B on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Olmo 3 1125 32B. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 65.3 GB of usable unified memory (RAM minus macOS overhead).
Can I run Olmo 3 1125 32B locally?
Yes — Olmo 3 1125 32B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Olmo 3 1125 32B?
At BF16, Olmo 3 1125 32B can reach ~74 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 65.3 × 0.65 = ~80 tok/s
Estimated speed at BF16 (65.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Olmo 3 1125 32B?
At BF16, the download is about 64.47 GB.
Which GPUs can run Olmo 3 1125 32B?
No single consumer GPU has enough VRAM to run Olmo 3 1125 32B at BF16 (65.3 GB). Multi-GPU or professional hardware is required.
Which devices can run Olmo 3 1125 32B?
30 devices with unified memory can run Olmo 3 1125 32B at BF16 (65.3 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.