Skip to content
Weights Not Walls
Menu

ScienceQA

Reported

Multiple-choice science questions from school curricula, many with an image.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Score shown
Epoch's column "Score", as a percentage, higher is better. The source gives no error bars.
Results
26 in all: 11 for open-weight models, 5 for closed ones, 10 where Epoch does not say. 7 linked to a model page here.

Best open-weight result: microsoft/Phi-3.5-vision-instruct, 91.3%, number 1 of 26 results.

ScienceQA results
#ModelWeightsSettingScoreDateSource
1microsoft/Phi-3.5-vision-instructEpoch: Phi-3.5-vision-instructopendefault91.3%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
2gpt-4o-2024-05-13OpenAIcloseddefault88.5%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
3gemini-1.0-pro-visionGoogle DeepMindcloseddefault79.7%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
4tiiuae/falcon-11B-vlmEpoch: falcon-11B-vlmopendefault74.9%not datedFalcon2-11B Technical Report
5Salesforce/blip2-opt-2.7bEpoch: blip2-opt-2.7b, Salesforce Researchopendefault74.2%not datedScienceQA leaderboard
6text-davinci-001OpenAIcloseddefault74.0%not datedScienceQA leaderboard
7text-davinci-001OpenAIcloseddefault74.0%not datedScienceQA leaderboard
8llama3-llava-next-8bnot stateddefault73.7%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
9llava-v1.6-vicuna-13bnot stateddefault73.6%not datedFalcon2-11B Technical Report
10llava-v1.6-mistral-7bnot stateddefault72.8%not datedFalcon2-11B Technical Report
11MM1-7B-Chatnot stateddefault72.6%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
12claude-3-haiku-20240307Anthropiccloseddefault72.0%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
13llava-v1.6-vicuna-7bnot stateddefault70.6%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
14InternVL-Chat-ViT-6B-Vicuna-13Bnot stateddefault70.1%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
15llava-v1.6-vicuna-7bnot stateddefault70.1%not datedFalcon2-11B Technical Report
16MM1-3B-Chatnot stateddefault69.4%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
17Qwen/Qwen-VL-ChatEpoch: Qwen-VL-Chatopendefault68.2%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
18Qwen/Qwen-VL-ChatEpoch: Qwen-VL-Chatopendefault67.2%not datedPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
19llava-v1.5-7bnot stateddefault66.8%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
20InternVL-Chat-ViT-6B-Vicuna-7Bnot stateddefault66.2%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
21Salesforce/instructblip-vicuna-13bEpoch: instructblip-vicuna-13bopendefault63.1%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
22Salesforce/instructblip-vicuna-7bEpoch: instructblip-vicuna-7bopendefault60.5%not datedVisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
23Llama-2-13bMeta AIno page matchopendefault55.8%not datedScienceQA leaderboard
24LLaMA-13BMeta AIno page matchopendefault43.3%not datedScienceQA leaderboard
25Llama-2-7bMeta AIno page matchopendefault43.1%not datedScienceQA leaderboard
26LLaMA-7BMeta AIno page matchopendefault36.2%not datedScienceQA leaderboard

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method