Edge Arena: Live & Deployed

Compare small AI models for edge deployment

Showing 6 of 6 models

TinyLlama 1.1B

Open Source

46.6%

Community-driven ultra-compact model trained on 3 trillion tokens, perfect for IoT and embedded systems.

1.1B
2,048 ctx
Q8_0
IoT Devices
MMLU
25.3%
HellaSwag
59.2%
ARC
55.4%
tinyiotedge
ollama run tinyllama

Gemma 2B

Google

62.3%

Google's ultra-efficient model designed specifically for mobile and resource-constrained deployments.

2B
8,192 ctx
Q5_K_M
Mobile
MMLU
42.3%
HellaSwag
71.4%
ARC
73.2%
mobilelightweightefficient
ollama run gemma:2b

Phi-3 Mini

Microsoft

76.8%

Microsoft's compact powerhouse optimized for complex reasoning tasks with exceptional performance-to-size ratio.

3.8B
4,096 ctx
Q4_K_M
Reasoning
MMLU
68.8%
HellaSwag
76.7%
ARC
84.9%
reasoninginstructioncompact
ollama run phi3

Mistral 7B

Mistral AI

74.0%

Mistral's breakthrough model with sliding window attention, excelling at creative and narrative tasks.

7B
32,768 ctx
Q4_K_M
Creative Writing
MMLU
62.5%
HellaSwag
81.3%
ARC
78.1%
creativelong-contextstorytelling
ollama run mistral

Qwen 1.5 7B

Alibaba

71.6%

Alibaba's multilingual model with exceptional mathematical reasoning and code generation abilities.

7B
32,768 ctx
Q4_K_M
Math/Code
MMLU
61.0%
HellaSwag
78.5%
ARC
75.2%
mathcodemultilingual
ollama run qwen:7b

Llama 3 8B

Meta

75.7%

Meta's flagship open model delivering strong all-around performance for diverse applications.

8B
8,192 ctx
Q4_K_M
General Purpose
MMLU
66.6%
HellaSwag
82.0%
ARC
78.6%
general-purposeversatilepopular
ollama run llama3