ScienceQA

Measures: Multiple-choice science questions from grade 1 to 12 curricula, many with an image or diagram, annotated with lectures and explanations.

Leaderboard

Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.

Leaderboard
Model	Accuracy	Provenance	Source	Evaluated
Phi-3.5-vision Microsoft	91.3%	Self-reported	Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone via Epoch AI	—
GPT-4o OpenAI	88.5%	Self-reported	Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone via Epoch AI	—
Gemini 1.0 Pro Vision Google DeepMind	79.7%	Self-reported	Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone via Epoch AI	—
Falcon 2 11B VLM Technology Innovation Institute (TII)	74.9%	Self-reported	Falcon2-11B Technical Report via Epoch AI	—
text-davinci-001 (InstructGPT) OpenAI	74.0%	Third party	ScienceQA leaderboard via Epoch AI	—
text-davinci-001 (InstructGPT) OpenAI	74.0%	Third party	ScienceQA leaderboard via Epoch AI	—
Claude 3 Haiku Anthropic	72.0%	Self-reported	Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone via Epoch AI	—
Qwen-VL-Chat Alibaba Group	68.2%	Third party	VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks via Epoch AI	—
Qwen-VL-Chat Alibaba Group	67.2%	Self-reported	Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone via Epoch AI	—
Llama 2 13B Meta	55.8%	Third party	ScienceQA leaderboard via Epoch AI	—
Llama 2 7B Meta	43.1%	Third party	ScienceQA leaderboard via Epoch AI	—