Model

OLMoE-1B-7B-0924

OPEN

Allen Institute for AI (Ai2) · OLMoE family · released Sep 3, 2024

First 100% open sparse MoE LLM (weights, data, code, logs); competitive with much larger dense models like Llama2-13B

ReasoningCodingVisionFunction callingTool useAgentic

104.8

Elo · rank #383

Parameters

6.9B

Active params

1.3B (MoE)

Context

4K tokens

Architecture

Sparse MoE (64 experts, 8 active/token, 16 layers)

License

Apache-2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning62.1%#79

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning84.2%#14

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning33.9%#121

best: ERNIE 4.5 300B-A47B · 94.3%

HellaSwagReasoning80.0%#84

best: Claude 3 Opus · 95.4%

IFEvalReasoning21.9%#152

best: Gemma 4 26B A4B · 98.5%

MMLU-ProKnowledge17.4%#173

best: Claude Fable 5 · 91.5%

MMLUKnowledge54.1%#226

best: OpenAI o3 · 92.9%

PIQAReasoning79.8%#48

best: GPT-4o mini · 93.1%

WinoGrandeReasoning70.2%#101

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

4.2 GB

VRAM @ FP16

13.8 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Only 1.3B active params make this fast on modest GPUs, though no verified RTX 4090 llama.cpp number was found

Quantizations

GGUF

OLMoE family

Elo progression across releases

OLMoE-1B-7B-0924Sep 2024104.8

Resources

HFallenai/OLMoE-1B-7B-0924 GHallenai/OLMoE DOCSallenai.org/olmo

API price weights · each benchmark row carries its own source badge (see methodology)