Atlas

Benchmarks

← All benchmarks

EQ-Bench v2 + MAGI-Hard Combined

Chat & Writing · 2024-04-01

This legacy leaderboard metric is the rounded arithmetic mean of separately reported EQ-Bench v2 and MAGI-Hard performance. It is retained as a diagnostic aggregate rather than an independent capability factor.

Top models (higher is better)

ModelScore
Llama 3.1 405B Instruct83.4
Gemini 1.5 Pro 00282.7
Claude 3.5 Sonnet (June 2024)82.6
GPT-4o (source unspecified)82.2
GPT-4 Turbo82.0
ChatGPT-4o Latest (September 2024 benchmark entry)81.5
GPT-4 061381.3
GPT-4 (0314)80.7
GPT-4 Turbo (1106 Preview)80.5
GPT-4 0125 Preview80.3
Llama 3.2 90B Vision Instruct79.9
Hermes 3 Llama 3.1 405B79.5
Opus 379.4
Mistral Large 2 (Instruct 2407)78.7
Qwen2 72B Instruct78.5
Qwen2.5 72B Instruct78.4
Qwen2.5 Instruct 32B78.3
Mistral Large 1.076.4
Qwen2.5 14B Instruct76.0
Llama 3 70B Instruct75.0
Qwen1.5 110B Chat74.9
Solar Pro Instruct Preview74.7
Smaug-Llama-3-70B-Instruct74.0
Qwen1.5 72B Chat73.1
Gemma 2 27B IT72.3
GPT-4o Mini72.2
Phi-3.5-MoE-instruct72.1
DeepSeek-V2.572.0
DeepSeek-V2-Chat-062871.9
Phi-3 Medium 4K Instruct71.4
Claude 3 Sonnet70.7
Mixtral 8x22B Instruct70.6
Qwen 72B Chat70.5
Smaug-72B-v0.170.0
Gemma 2 9B IT69.2
Yi 1.5 34B Chat68.9
Phi-3-Small-8K-Instruct68.8
WizardLM-2 8x22B68.5
Quyen Pro Max V0.168.2
Qwen1.5 32B Chat68.2
Loading Atlas data…