Model

Orca 2 (13B)

OPEN

Microsoft · Orca family · released Nov 18, 2023

Reasoning-focused Llama-2 13B fine-tune trained with 'Prompt Erasure' to select its own reasoning strategy; rivals models 5-10x larger.

ReasoningCodingVisionFunction callingTool useAgentic

292.2

Elo · rank #348

Parameters

13B

Active params

13B (dense)

Context

4K tokens

Architecture

Dense decoder-only Transformer (Llama 2 13B backbone)

License

Microsoft Research License

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning49.9%#19

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning83.4%#38

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning92.8%#4

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning50.2%#99

best: ERNIE 4.5 300B-A47B · 94.3%

DROPReasoning57.97#53

best: Hunyuan-T1 · 93.1

GSM8KMath59.1%#117

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning77.8%#93

best: Claude 3 Opus · 95.4%

LAMBADAReasoning63.7%#25

best: GPT-3 175B · 86.4%

MMLUKnowledge57.7%#214

best: OpenAI o3 · 92.9%

MT-BenchHuman preference6.15#65

best: Hunyuan-Large (A52B) · 9.4

Run it locally

VRAM @ Q4

7.5 GB

VRAM @ FP16

26 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 · GPTQ · AWQ

Orca family

Elo progression across releases

Orca 2 (13B)Nov 2023292.2

Resources

HFmicrosoft/Orca-2-13b GHmicrosoft/orca

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