tinyanalysis

Leaderboard / Ai2 (OLMo)

OLMo 3.1 32B Think

Ai2 (OLMo) · OLMo 3 · released 2025-12-10 · data checked 2026-11-01

OLMo 3.1 32B Think is a 32.2B-parameter dense model from Ai2 (OLMo) (United States), released 2025-12-10 under the Apache-2.0 license. It scores 57.0 on the Small LM Index, #22 of 34 medium models. It needs about 22 GB of memory at 4-bit with an 8K-token context.

Small LM Index
57.0
#22 of 34 scored medium models
Likely 52.6–61.2 · from 3 of 6 core benchmarks
Parameters
32.2B
Dense
Price per 1M tokens
—
No hosted API on OpenRouter; self-host
Context window
64K
tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 6 reported by Ai2 (OLMo). How we label sources.

GPQA Diamond Lab-reported Knowledge57.5
Class best: Nanbeige4.3 33B 89.5
AIME Lab-reported Math · AIME 202578.1
Class best: Ternary Bonsai 2 27B 95.8
LiveCodeBench Lab-reported Coding · LiveCodeBench v383.3
Class best: Qwen3.8 27B 90.3
IFBench Lab-reported Instruction following68.1
Class best: Sage-A1.5 81.6
MMLU Lab-reported Knowledge86.4
Class best: EXAONE 4.0 32B 92.3
IFEval Lab-reported Instruction following93.8
Class best: LFM2.5 24B-A2B 96.8

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 77
Humanity's Last Exam≈ 9
SWE-bench Verified≈ 41

Specifications

Lab
Ai2 (OLMo) (US · United States)
Architecture
Dense transformer
Total parameters
32.2B
Active parameters
All (dense)
Size class
Medium
Context window
64K tokens
Input modalities
text
Reasoning
Reasoning (thinking)
License
Apache-2.0
Openness
Fully open: weights, training data and code
Memory to run
~22 GB at 4-bit, 8K context (more sizes)
Hugging Face
6.9K downloads (30 days) · 119 likes
Weights on Hugging Face ↗

Compare with

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses unsloth/Olmo-3.1-32B-Think-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_NL, IQ4_XS, Q2_K, Q2_K_L, Q3_K_M, Q3_K_S, Q4_0….

Ollama
ollama run hf.co/unsloth/Olmo-3.1-32B-Think-GGUF:Q4_K_M
llama.cpp
llama-server -hf unsloth/Olmo-3.1-32B-Think-GGUF:Q4_K_M
Then open http://localhost:8080 for a chat window.
MLX (Mac)
pip install -U mlx-lm
mlx_lm.chat --model mlx-community/Olmo-3.1-32B-Think-4bit
vLLM (GPU server)
vllm serve allenai/Olmo-3.1-32B-Think
Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization4K context32K context64K context
16-bit F1667 GB68 GB71 GB
8-bit Q8_036 GB38 GB40 GB
4-bit Q4_K_M22 GB24 GB26 GB

Check it against your GPU, Mac or phone →