Skip to content
Weights Not Walls
Menu

CSQA2

Reported

No description yet: we only describe an eval set when we can do it correctly.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://arxiv.org/pdf/2201.05320
Score shown
Epoch's column "Score", as a percentage, higher is better. The source gives no error bars.
Results
6 in all: 3 for open-weight models, 2 for closed ones, 1 where Epoch does not say. 1 linked to a model page here.

Best open-weight result: T5-11B, 67.8%, number 1 of 6 results.

CSQA2 results
#ModelWeightsSettingScoreDateSource
1T5-11BGoogleno page matchopendefault67.8%not datedCommonsenseQA 2.0: Exposing the Limits of AI through Gamification
2T5-3BGoogleno page matchopendefault60.2%not datedDecker: Double Check with Heterogeneous Knowledge for Commonsense Fact Verification
3gpt-3.5-turbo-0613OpenAIclosedShots: few57.0%not datedarxiv.org
4T5-Largenot stateddefault54.6%not datedCommonsenseQA 2.0: Exposing the Limits of AI through Gamification
5text-davinci-001OpenAIcloseddefault52.9%not datedCommonsenseQA 2.0: Exposing the Limits of AI through Gamification
6meta-llama/Llama-2-70b-hfEpoch: Llama-2-70b-hf, Meta AIopenShots: few50.0%not datedarxiv.org

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