Model

Skywork-o1-Open-Llama-3.1-8B

OPEN

Kunlun Tech · Skywork o1 family · released Nov 26, 2024

Early open o1-style reasoning model built on Llama-3.1-8B with PRM-guided slow-thinking reasoning traces

ReasoningCodingVisionFunction callingTool useAgentic

987.8

Elo · rank #215

Parameters

Active params

8B (dense)

Context

128K tokens

Architecture

Dense transformer (Llama-3.1-8B architecture), fine-tuned with o1-style slow-thinking + process-reward-model data

License

Skywork Community License

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AMCMath55.0%#3

best: Falcon-H1 34B · 69.4%

Gaokao MathMath63.6%#3

best: Doubao-Seed-1.6 · 96.0%

GSM8KMath91.6%#34

best: Llama 3.1 405B · 96.8%

HumanEval+Coding78.7%#16

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding82.3%#50

best: Claude Opus 4.5 · 99.4%

LiveCodeBenchCoding26.0%#158

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

MATH-500Math78.1%#71

best: GPT-5 · 99.4%

MBPP+Coding65.2%#22

best: Llama 3.1 405B · 88.6%

MBPPCoding77.9%#19

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

OlympiadBenchMath43.1%#2

best: Doubao-1.5-Pro · 59.8%

Run it locally

VRAM @ Q4

4.92 GB

VRAM @ FP16

16.1 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_K_M

Skywork o1 family

Elo progression across releases

Skywork-o1-Open-Llama-3.1-8BNov 2024987.8

Resources

HFSkywork/Skywork-o1-Open-Llama-3.1-8B GHSkyworkAI/skywork-o1-prm-inference

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