Skip to content

Olmo 3 32B Think VRAM requirements

How much VRAM Olmo 3 32B Think needs
4-bit weights: 18.2 GBWith 8K context: 20.2 GBWith 32K context: 21.7 GB8-bit, 8K context: 33.9 GB
Parameters
32.2B
Longest context
64K
Released
Nov 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Olmo 3 32B Think needs

Memory Olmo 3 32B Think needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 64K
16-bit60.0 GB62.1 GB63.6 GB65.6 GB
8-bit31.9 GB33.9 GB35.4 GB37.4 GB
6-bit24.6 GB26.7 GB28.2 GB30.2 GB
5-bit21.4 GB23.4 GB24.9 GB26.9 GB
4-bit18.2 GB20.2 GB21.7 GB23.7 GB
3-bit14.7 GB16.7 GB18.2 GB20.2 GB
2-bit11.3 GB13.3 GB14.8 GB16.8 GB
1.6-bit6.2 GB8.2 GB9.7 GB11.7 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Olmo 3 32B Think

The best way Olmo 3 32B Think runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Olmo 3 32B Think on RTX 4060 report
RTX 5060Too bigPDF: Olmo 3 32B Think on RTX 5060 report
RTX 3070Too bigPDF: Olmo 3 32B Think on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Olmo 3 32B Think on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Olmo 3 32B Think on RTX 3060 12 GB report
Arc B580Too bigPDF: Olmo 3 32B Think on Arc B580 report
RTX 4070Too bigPDF: Olmo 3 32B Think on RTX 4070 report
RTX 4070 SuperToo bigPDF: Olmo 3 32B Think on RTX 4070 Super report
RTX 5070Too bigPDF: Olmo 3 32B Think on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Olmo 3 32B Think on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Olmo 3 32B Think on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Olmo 3 32B Think on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Olmo 3 32B Think on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Olmo 3 32B Think on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Olmo 3 32B Think on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Olmo 3 32B Think on RTX 5070 Ti report
RTX 5080Too bigPDF: Olmo 3 32B Think on RTX 5080 report
Radeon RX 7900 XTX4-bit48K27 tok/s3.1 sPDF: Olmo 3 32B Think on Radeon RX 7900 XTX report
RTX 30904-bit48K32 tok/s2.0 sPDF: Olmo 3 32B Think on RTX 3090 report
RTX 40904-bit48K34 tok/s0.8 sPDF: Olmo 3 32B Think on RTX 4090 report
Radeon AI PRO R97006-bit64K14 tok/s2.0 sPDF: Olmo 3 32B Think on Radeon AI PRO R9700 report
RTX 50906-bit64K44 tok/s0.7 sPDF: Olmo 3 32B Think on RTX 5090 report
2× RTX 30908-bit64K19 tok/s1.1 sPDF: Olmo 3 32B Think on 2× RTX 3090 report
2× RTX 40908-bit64K20 tok/s0.5 sPDF: Olmo 3 32B Think on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit64K26 tok/s1.0 sPDF: Olmo 3 32B Think on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit64K19 tok/s0.7 sPDF: Olmo 3 32B Think on RTX 6000 Ada report
L40S8-bit64K17 tok/s1.2 sPDF: Olmo 3 32B Think on L40S report
2× RTX 50908-bit64K35 tok/s0.4 sPDF: Olmo 3 32B Think on 2× RTX 5090 report
4× RTX 30908-bit64K19 tok/s0.6 sPDF: Olmo 3 32B Think on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit64K35 tok/s0.5 sPDF: Olmo 3 32B Think on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Olmo 3 32B Think on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Olmo 3 32B Think on Mac mini M4 24 GB report
Mac mini M4 32 GB3-bit64K6.1 tok/s44 sPDF: Olmo 3 32B Think on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Olmo 3 32B Think on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit64K8.2 tok/s22 sPDF: Olmo 3 32B Think on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit64K6.4 tok/s22 sPDF: Olmo 3 32B Think on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit64K11 tok/s11 sPDF: Olmo 3 32B Think on M4 Max 128 GB report
M5 Max 128 GB8-bit64K13 tok/s3.2 sPDF: Olmo 3 32B Think on M5 Max 128 GB report
M3 Ultra 96 GB8-bit64K15 tok/s6.2 sPDF: Olmo 3 32B Think on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit64K15 tok/s6.2 sPDF: Olmo 3 32B Think on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit64K15 tok/s6.2 sPDF: Olmo 3 32B Think on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit64K21 tok/s1.9 sPDF: Olmo 3 32B Think on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit64K4.5 tok/s6.4 sPDF: Olmo 3 32B Think on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit64K4.5 tok/s6.4 sPDF: Olmo 3 32B Think on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit64K6.2 tok/s2.2 sPDF: Olmo 3 32B Think on RTX Spark report
DGX Spark8-bit64K5.6 tok/s2.2 sPDF: Olmo 3 32B Think on DGX Spark report
Dual DGX Spark8-bit64K5.6 tok/s1.3 sPDF: Olmo 3 32B Think on Dual DGX Spark report
DGX Station GB3008-bit64K115 tok/s0.2 sPDF: Olmo 3 32B Think on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit64K35 tok/s0.3 sPDF: Olmo 3 32B Think on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit64K35 tok/s0.2 sPDF: Olmo 3 32B Think on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit64K35 tok/s0.1 sPDF: Olmo 3 32B Think on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit64K39 tok/s1.4 sPDF: Olmo 3 32B Think on A100 80 GB SXM report
H100 SXM8-bit64K61 tok/s0.4 sPDF: Olmo 3 32B Think on H100 SXM report
4× H200 NVL8-bit64K84 tok/s0.2 sPDF: Olmo 3 32B Think on 4× H200 NVL report
8× H200 NVL8-bit64K84 tok/s0.1 sPDF: Olmo 3 32B Think on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/Olmo-3-32B-Think-GGUF
vLLM
vllm serve allenai/Olmo-3-32B-Think

Questions

How much VRAM does Olmo 3 32B Think need?

Olmo 3 32B Think needs 18.2 GB for its weights at 4-bit, 20.2 GB with an 8K context and 21.7 GB with 32K. At 8-bit, 33.9 GB.

What is the minimum VRAM for Olmo 3 32B Think?

About 16.7 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the Radeon RX 7900 XTX.

How big is Olmo 3 32B Think?

32.2B parameters: 60.0 GB of weights at 16-bit, 18.2 GB at 4-bit.

Can Olmo 3 32B Think run on an RTX 4090?

Yes. At 4-bit it holds 48K of context and writes about 34 tokens a second.

More models