Model

OLMo 3 7B

OPEN

Allen Institute for AI (Ai2) · OLMo 3 family · released Nov 20, 2025

Lighter-weight member of the OLMo 3 family (Base/Think/Instruct) built on the Dolma 3 / Dolci stack; runs on a wider range of hardware.

ReasoningCodingVisionFunction callingTool useAgentic

843.3

Elo · rank #258

Parameters

7.3B

Active params

7.3B (dense)

Context

64K tokens

Architecture

Dense transformer (decoder-only, sliding/full attention mix)

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

AGIEvalReasoning64.4%#5

best: OLMo 3-Think 32B · 88.2%

AIMEMath44.3%#123

best: GPT-5.2 · 100.0%

AlpacaEvalHuman preference40.9%#26

best: Tülu 2+DPO 70B · 95.1%

BIG-Bench HardReasoning71.2%#50

best: ERNIE 4.5 300B-A47B · 94.3%

BFCL v3Agents49.8%#30

best: Hunyuan-A13B · 78.3%

GPQA DiamondReasoning40.0%#226

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning5.8%#120

best: Claude Opus 5 · 64.7%

HumanEval+Coding77.2%#17

best: Mistral Small 3.2 (24B) · 92.9%

IFBenchReasoning32.3%#70

best: MiniMax M3 · 83.0%

IFEvalReasoning85.6%#56

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding29.5%#153

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math87.3%#55

best: GPT-5 · 99.4%

MBPP+Coding60.2%#29

best: Llama 3.1 405B · 88.6%

MMLU-ProKnowledge52.2%#129

best: Claude Fable 5 · 91.5%

MMLUKnowledge69.1%#152

best: OpenAI o3 · 92.9%

Run it locally

VRAM @ Q4

5 GB

VRAM @ FP16

15 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

Throughput data unavailable.

Quantizations

GGUF Q4

Fine-tune it

Permissive

QLoRA6.6 GB1× RTX 3060 12GB

LoRA17.2 GB1× RTX 3090 24GB

Full fine-tune118.8 GB1× H200 141GB

QLoRA SFT on ~10k samples ≈ $4.27 (1× RTX 3060 12GB)

Plan a fine-tune →

OLMo 3 family

Elo progression across releases

OLMo 3 7BNov 2025843.3 OLMo 3-Think 32BNov 20251448.0

Resources

HFallenai/Olmo-3-7B-Instruct GHallenai/OLMo-core DOCSallenai.org/blog/olmo3 APIAvailable via Allen Institute for AI (Ai2) API

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