PIQA
ReportedPhysical common sense: choosing the better of two ways to do an everyday task.
- 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; random guessing scores about 50.0%. The source gives no error bars.
- Results
- 113 in all: 82 for open-weight models, 27 for closed ones, 4 where Epoch does not say. 25 linked to a model page here.
Best open-weight result: microsoft/Phi-3.5-MoE-instruct, 88.6%, number 2 of 113 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | gpt-4o-mini-2024-07-18OpenAI | closed | Shots: 5 | 88.7% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 2 | microsoft/Phi-3.5-MoE-instructEpoch: Phi-3.5-MoE-instruct, Microsoft | open | Shots: 5 | 88.6% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 3 | gemini-1.5-flash-002Google DeepMind | closed | Shots: 5 | 87.5% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 4 | Llama-3.1-405BMeta AIno page match | open | Shots: 0 | 85.9% | not dated | DeepSeek-V3 Technical Report |
| 5 | falcon-180BTechnology Innovation Instituteno page match | open | default | 84.9% | not dated | The Falcon Series of Open Language Models |
| 6 | deepseek-ai/DeepSeek-V3-BaseEpoch: DeepSeek-V3-Base, DeepSeek | open | Shots: 0 | 84.7% | not dated | DeepSeek-V3 Technical Report |
| 7 | Inflection-1Inflection AI | closed | default | 84.2% | not dated | The Falcon Series of Open Language Models |
| 8 | DeepSeek-V2DeepSeekno page match | open | Shots: 0 | 83.9% | not dated | DeepSeek-V3 Technical Report |
| 9 | gemma-2-9bGoogle DeepMindno page match | open | Shots: 5 | 83.7% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 10 | mistralai/Mixtral-8x7B-v0.1Epoch: Mixtral-8x7B-v0.1, Mistral AI | open | default | 83.6% | not dated | Mixtral of Experts |
| 11 | mistralai/Mistral-Nemo-Base-2407Epoch: Mistral-Nemo-Base-2407, Mistral AI | open | Shots: 5 | 83.5% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 12 | stabilityai/StableBeluga2Epoch: StableBeluga2, Stability AI | open | Shots: 0 | 83.3% | not dated | Qwen Technical Report |
| 13 | Megatron-Turing NLG 530BMicrosoft,Nvidia | closed | Shots: few | 83.2% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 14 | falcon-40bTechnology Innovation Instituteno page match | open | default | 83.0% | not dated | The Falcon Series of Open Language Models |
| 15 | mistralai/Mistral-7B-v0.1Epoch: Mistral-7B-v0.1, Mistral AI | open | default | 83.0% | not dated | Nemotron-4 15B Technical Report |
| 16 | meta-llama/Llama-2-70b-hfEpoch: Llama-2-70b-hf, Meta AI | open | Shots: 0 | 82.8% | not dated | Qwen Technical Report |
| 17 | meta-llama/Llama-2-70b-hfEpoch: Llama-2-70b-hf, Meta AI | open | default | 82.8% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 18 | meta-llama/Llama-2-70b-hfEpoch: Llama-2-70b-hf, Meta AI | open | default | 82.8% | not dated | The Falcon Series of Open Language Models |
| 19 | LLaMA-65BMeta AIno page match | open | Shots: 0 | 82.8% | not dated | Qwen Technical Report |
| 20 | LLaMA-65BMeta AIno page match | open | default | 82.8% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 21 | LLaMA-65BMeta AIno page match | open | default | 82.8% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 22 | meta-llama/Llama-2-70b-hfEpoch: Llama-2-70b-hf, Meta AI | open | default | 82.6% | not dated | Mixtral of Experts |
| 23 | Qwen2.5-72BAlibabano page match | open | Shots: 0 | 82.6% | not dated | DeepSeek-V3 Technical Report |
| 24 | falcon-40bTechnology Innovation Instituteno page match | open | default | 82.4% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 25 | Nemotron-4 15BNvidia | closed | default | 82.4% | not dated | Nemotron-4 15B Technical Report |
| 26 | LLaMA-33BMeta AIno page match | open | Shots: 0 | 82.3% | not dated | Qwen Technical Report |
| 27 | LLaMA-33BMeta AIno page match | open | default | 82.3% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 28 | LLaMA-33BMeta AIno page match | open | default | 82.3% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 29 | PaLM 540BGoogle Research | closed | default | 82.3% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 30 | PaLM 540BGoogle Research | closed | default | 82.3% | not dated | The Falcon Series of Open Language Models |
| 31 | text-davinci-001OpenAI | closed | Shots: few | 82.3% | not dated | GLaM: Efficient Scaling of Language Models with Mixture-of-Experts |
| 32 | text-davinci-001OpenAI | closed | Shots: few | 82.3% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 33 | LLaMA-33BMeta AIno page match | open | default | 82.2% | not dated | Mixtral of Experts |
| 34 | mistralai/Mistral-7B-Instruct-v0.1Epoch: Mistral-7B-Instruct-v0.1, Mistral AI | open | default | 82.2% | not dated | Mixtral of Experts |
| 35 | mistralai/Mistral-7B-Instruct-v0.2Epoch: Mistral-7B-Instruct-v0.2, Mistral AI | open | Shots: 0 | 82.2% | not dated | Gemma: Open Models Based on Gemini Research and Technology |
| 36 | Megatron-Turing NLG 530BMicrosoft,Nvidia | closed | default | 82.0% | not dated | The Falcon Series of Open Language Models |
| 37 | Megatron-Turing NLG 530BMicrosoft,Nvidia | closed | Shots: 0 | 82.0% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 38 | Llama-2-34bMeta AI | closed | default | 81.9% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 39 | Llama-2-34bMeta AI | closed | default | 81.9% | not dated | Nemotron-4 15B Technical Report |
| 40 | Llama-2-34bMeta AI | closed | default | 81.9% | not dated | The Falcon Series of Open Language Models |
| 41 | mpt-30bMosaicMLno page match | open | default | 81.9% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 42 | Chinchilla (70B)DeepMind | closed | default | 81.8% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 43 | Chinchilla (70B)DeepMind | closed | default | 81.8% | not dated | The Falcon Series of Open Language Models |
| 44 | Gopher (280B)DeepMind | closed | Shots: 0 | 81.8% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 45 | Gopher (280B)DeepMind | closed | Shots: 0 | 81.8% | not dated | Scaling Language Models: Methods, Analysis & Insights from Training Gopher |
| 46 | Gopher (280B)DeepMind | closed | default | 81.8% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 47 | Gopher (280B)DeepMind | closed | default | 81.8% | not dated | The Falcon Series of Open Language Models |
| 48 | gemma-7bGoogle DeepMindno page match | open | Shots: 0 | 81.2% | not dated | Gemma: Open Models Based on Gemini Research and Technology |
| 49 | gemma-7bGoogle DeepMindno page match | open | default | 81.2% | not dated | Nemotron-4 15B Technical Report |
| 50 | meta-llama/Llama-3.1-8B-InstructEpoch: Llama-3.1-8B-Instruct, Meta AI | open | Shots: 5 | 81.2% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 51 | microsoft/Phi-3.5-mini-instructEpoch: Phi-3.5-mini-instruct, Microsoft | open | Shots: 5 | 81.0% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 52 | text-davinci-001OpenAI | closed | Shots: 0 | 81.0% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 53 | text-davinci-001OpenAI | closed | default | 81.0% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 54 | text-davinci-001OpenAI | closed | default | 81.0% | not dated | The Falcon Series of Open Language Models |
| 55 | Megatron-Turing NLG 530BMicrosoft,Nvidia | closed | Shots: 1 | 81.0% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 56 | Llama-2-13bMeta AIno page match | open | default | 80.8% | not dated | Mixtral of Experts |
| 57 | mpt-7bMosaicMLno page match | open | Shots: 0 | 80.6% | not dated | Qwen Technical Report |
| 58 | mpt-7bMosaicMLno page match | open | default | 80.6% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 59 | Llama-2-13bMeta AIno page match | open | Shots: 0 | 80.5% | not dated | Gemma: Open Models Based on Gemini Research and Technology |
| 60 | Llama-2-13bMeta AIno page match | open | Shots: 0 | 80.5% | not dated | Qwen Technical Report |
| 61 | Llama-2-13bMeta AIno page match | open | default | 80.5% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 62 | Llama-2-13bMeta AIno page match | open | default | 80.5% | not dated | The Falcon Series of Open Language Models |
| 63 | PaLM 62B | not stated | default | 80.5% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 64 | text-davinci-001OpenAI | closed | Shots: 1 | 80.5% | not dated | GLaM: Efficient Scaling of Language Models with Mixture-of-Experts |
| 65 | text-davinci-001OpenAI | closed | Shots: 1 | 80.5% | not dated | Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model |
| 66 | text-davinci-001OpenAI | closed | Shots: 0 | 80.4% | not dated | GLaM: Efficient Scaling of Language Models with Mixture-of-Experts |
| 67 | falcon-7bTechnology Innovation Instituteno page match | open | default | 80.3% | not dated | The Falcon Series of Open Language Models |
| 68 | internlm/internlm-20bEpoch: internlm-20b | open | Shots: 0 | 80.3% | not dated | Qwen Technical Report |
| 69 | LLaMA-13BMeta AIno page match | open | Shots: 0 | 80.1% | not dated | Qwen Technical Report |
| 70 | LLaMA-13BMeta AIno page match | open | default | 80.1% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 71 | LLaMA-13BMeta AIno page match | open | default | 80.1% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 72 | Qwen-14BAlibabano page match | open | Shots: 0 | 79.9% | not dated | Qwen Technical Report |
| 73 | Qwen-14BAlibabano page match | open | default | 79.9% | not dated | Nemotron-4 15B Technical Report |
| 74 | Llama-2-13bMeta AIno page match | open | default | 79.8% | not dated | Nemotron-4 15B Technical Report |
| 75 | LLaMA-7BMeta AIno page match | open | Shots: 0 | 79.8% | not dated | Qwen Technical Report |
| 76 | LLaMA-7BMeta AIno page match | open | default | 79.8% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 77 | LLaMA-7BMeta AIno page match | open | default | 79.8% | not dated | LLaMA: Open and Efficient Foundation Language Models |
| 78 | falcon-7bTechnology Innovation Instituteno page match | open | default | 79.4% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 79 | falcon-7bTechnology Innovation Instituteno page match | open | Shots: 0 | 79.4% | not dated | XGen-7B Technical Report |
| 80 | mpt-7bMosaicMLno page match | open | Shots: 0 | 79.1% | not dated | XGen-7B Technical Report |
| 81 | mpt-7bMosaicMLno page match | open | default | 78.9% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 82 | Llama-2-7bMeta AIno page match | open | Shots: 0 | 78.8% | not dated | Gemma: Open Models Based on Gemini Research and Technology |
| 83 | Llama-2-7bMeta AIno page match | open | Shots: 0 | 78.8% | not dated | Qwen Technical Report |
| 84 | Llama-2-7bMeta AIno page match | open | default | 78.8% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 85 | Llama-2-7bMeta AIno page match | open | default | 78.8% | not dated | The Falcon Series of Open Language Models |
| 86 | LLaMA-7BMeta AIno page match | open | Shots: 0 | 78.7% | not dated | XGen-7B Technical Report |
| 87 | Baichuan-2-13B-BaseBaichuanno page match | open | Shots: 0 | 78.1% | not dated | Qwen Technical Report |
| 88 | Baichuan-2-13B-BaseBaichuanno page match | open | default | 78.1% | not dated | Nemotron-4 15B Technical Report |
| 89 | Llama-2-7bMeta AIno page match | open | default | 78.1% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 90 | internlm/internlm-7bEpoch: internlm-7b | open | Shots: 0 | 77.9% | not dated | Qwen Technical Report |
| 91 | Llama-2-7bMeta AIno page match | open | default | 77.9% | not dated | Mixtral of Experts |
| 92 | LLaMA-7BMeta AIno page match | open | default | 77.9% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 93 | Qwen-7BAlibabano page match | open | Shots: 0 | 77.9% | not dated | Qwen Technical Report |
| 94 | vicuna-13b-v1.1 | not stated | default | 77.4% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 95 | gemma-2bGoogle DeepMindno page match | open | Shots: 0 | 77.3% | not dated | Gemma: Open Models Based on Gemini Research and Technology |
| 96 | RedPajama-INCITE-7B-Base | not stated | Shots: 0 | 76.9% | not dated | XGen-7B Technical Report |
| 97 | falcon-7bTechnology Innovation Instituteno page match | open | Shots: 0 | 76.7% | not dated | Qwen Technical Report |
| 98 | falcon-7bTechnology Innovation Instituteno page match | open | default | 76.7% | not dated | Llama 2: Open Foundation and Fine-Tuned Chat Models |
| 99 | EleutherAI/gpt-neox-20bEpoch: gpt-neox-20b, EleutherAI | open | Shots: 0 | 76.7% | not dated | XGen-7B Technical Report |
| 100 | baichuan-inc/Baichuan-7BEpoch: Baichuan-7B, Baichuan | open | Shots: 0 | 76.2% | not dated | Qwen Technical Report |
| 101 | open_llama_7b | not stated | Shots: 0 | 76.0% | not dated | XGen-7B Technical Report |
| 102 | facebook/opt-13bEpoch: opt-13b | open | Shots: 0 | 75.7% | not dated | XGen-7B Technical Report |
| 103 | Salesforce/xgen-7b-8k-baseEpoch: xgen-7b-8k-base, Salesforce | open | Shots: 0 | 75.5% | not dated | XGen-7B Technical Report |
| 104 | dolly-v2-12bDatabricksno page match | open | Shots: 0 | 75.4% | not dated | XGen-7B Technical Report |
| 105 | gpt-j-6bEleutherAI,LAIONno page match | open | Shots: 0 | 75.4% | not dated | XGen-7B Technical Report |
| 106 | Cerebras-GPT-13BCerebras Systemsno page match | open | Shots: 0 | 73.5% | not dated | XGen-7B Technical Report |
| 107 | mistralai/Mistral-7B-v0.1Epoch: Mistral-7B-v0.1, Mistral AI | open | Shots: 5 | 73.4% | not dated | Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone |
| 108 | Qwen/Qwen-1_8BEpoch: Qwen-1_8B | open | Shots: 0 | 73.3% | not dated | Qwen Technical Report |
| 109 | gpt-neo-2.7BEleutherAIno page match | open | default | 72.9% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 110 | gpt2-xlOpenAIno page match | open | default | 70.5% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 111 | zai-org/chatglm2-6bEpoch: chatglm2-6b | open | Shots: 0 | 69.6% | not dated | Qwen Technical Report |
| 112 | facebook/opt-1.3bEpoch: opt-1.3b, Meta AI | open | default | 69.0% | not dated | Textbooks Are All You Need II: phi-1.5 technical report |
| 113 | stabilityai/stablelm-tuned-alpha-7bEpoch: stablelm-tuned-alpha-7b | open | Shots: 0 | 65.8% | not dated | XGen-7B Technical Report |
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