BoolQ
Models read a Wikipedia passage and answer a naturally occurring yes or no question about it. Scored by accuracy.
Part of SuperGLUE. The paper reports 90 percent human accuracy, which large models now approach.
In the authors’ words
In this paper we study yes/no questions that are naturally occurring — meaning that they are generated in unprompted and unconstrained settings. We build a reading comprehension dataset, BoolQ, of such questions, and show that they are unexpectedly challenging. They often query for complex, non-factoid information, and require difficult entailment-like inference to solve. We also explore the effectiveness of a range of transfer learning baselines. We find that transferring from entailment data is more effective than transferring from paraphrase or extractive QA data, and that it, surprisingly, continues to be very beneficial even when starting from massive pre-trained language models such as BERT. Our best method trains BERT on MultiNLI and then re-trains it on our train set. It achieves 80.4% accuracy compared to 90% accuracy of human annotators (and 62% majority-baseline), leaving a significant gap for future work.
- Level
- Everyday Most adults, without training
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Multiple choice
PaLM 2-L leads with 90.9% among models released by Jun 2025
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Jun 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 90.9 PaLM 2-L
- 87.9 ChatGLM3-6B
- 86.6
phi-3-medium 14B - 85.0
Llama 2-70B - 84.8
Gemma 2 27B - 84.8 phi-3-small 7.4B
- 84.6 Phi-3.5-MoE
- 84.2 Gemma 2 9B
- 83.7 Llama 2-34B
- 83.3 Phi-2
- 83.2 Gemma 7B
- 82.4 Gemma 3 27B
- 81.7 Llama 2-13B
- 81.6 Gemma 3n E4B
- 81.2 Phi-4 Mini
- 80.2 BitNet b1.58 2B4T
- 80.0 Llama 3.1-405B
- 79.4 Llama 3.1-70B
- 79.0 Llama 3-70B
- 78.8 Gemma 3 12B
- 78.0 Phi-3.5-mini
- 77.4 Llama 2-7B
- 77.2 phi-3-mini 3.8B
- 75.7 Llama 3-8B
- 75.0 Llama 3.1-8B
- 72.3 Gemma 3 4B
- 69.4 Gemma 2B
- 63.2 Gemma 3 1B
- 1 PaLM 2-L 90.9%
- 2 ChatGLM3-6B 87.9%
- 3 phi-3-medium 14B 86.6%
- 4 Llama 2-70B 85%
- 5 Gemma 2 27B 84.8%
- 6 phi-3-small 7.4B 84.8%
- 7 Phi-3.5-MoE 84.6%
- 8 Gemma 2 9B 84.2%
- 9 Llama 2-34B 83.7%
- 10 Phi-2 83.3%
- 11 Gemma 7B 83.2%
- 12 Gemma 3 27B 82.4%
- 13 Llama 2-13B 81.7%
- 14 Gemma 3n E4B 81.6%
- 15 Phi-4 Mini 81.2%
- 16 BitNet b1.58 2B4T 80.2%
- 17 Llama 3.1-405B 80%
- 18 Llama 3.1-70B 79.4%
- 19 Llama 3-70B 79%
- 20 Gemma 3 12B 78.8%
- 21 Phi-3.5-mini 78%
- 22 Llama 2-7B 77.4%
- 23 phi-3-mini 3.8B 77.2%
- 24 Llama 3-8B 75.7%
- 25 Llama 3.1-8B 75%
- 26 Gemma 3 4B 72.3%
- 27 Gemma 2B 69.4%
- 28 Gemma 3 1B 63.2%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | PaLM 2-L | 90.9% | May 2023 | Model card | |
| 2 | ChatGLM3-6B | 87.9% | Jun 2024 | Model card | |
| 3 | phi-3-medium 14B | 86.6% | Apr 2024 | Model card | |
| 4 | Llama 2-70B | 85% | Jul 2023 | Model card | |
| 5 | Gemma 2 27B | 84.8% | Jun 2024 | Model card | |
| 6 | phi-3-small 7.4B | 84.8% | Aug 2024 | Model card | |
| 7 | Phi-3.5-MoE | 84.6% | Aug 2024 | Model card | |
| 8 | Gemma 2 9B | 84.2% | Jul 2024 | Model card | |
| 9 | Llama 2-34B | 83.7% | Jul 2023 | Model card | |
| 10 | Phi-2 | 83.3% | Dec 2023 | Model card | |
| 11 | Gemma 7B | 83.2% | Feb 2024 | Model card | |
| 12 | Gemma 3 27B | 82.4% | Mar 2025 | Model card | |
| 13 | Llama 2-13B | 81.7% | Jul 2023 | Model card | |
| 14 | Gemma 3n E4B | 81.6% | Jun 2025 | Model card | |
| 15 | Phi-4 Mini | 81.2% | Feb 2025 | Model card | |
| 16 | BitNet b1.58 2B4T | 80.2% | Apr 2025 | Model card | |
| 17 | Llama 3.1-405B | 80% | Jul 2024 | Model card | |
| 18 | Llama 3.1-70B | 79.4% | Jul 2024 | Model card | |
| 19 | Llama 3-70B | 79% | Apr 2024 | Model card | |
| 20 | Gemma 3 12B | 78.8% | Mar 2025 | Model card | |
| 21 | Phi-3.5-mini | 78% | Aug 2024 | Model card | |
| 22 | Llama 2-7B | 77.4% | Jul 2023 | Model card | |
| 23 | phi-3-mini 3.8B | 77.2% | Aug 2024 | Model card | |
| 24 | Llama 3-8B | 75.7% | Apr 2024 | Model card | |
| 25 | Llama 3.1-8B | 75% | Jul 2024 | Model card | |
| 26 | Gemma 3 4B | 72.3% | Mar 2025 | Model card | |
| 27 | Gemma 2B | 69.4% | Mar 2024 | Model card | |
| 28 | Gemma 3 1B | 63.2% | Mar 2025 | Model card |
The benchmark's own leaderboard has 26 entries, 3 of them models we track. Epoch AI has scores for 75 models. We plot every source in the charts below.
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard, Epoch AI’s runs and what labs claim in their cards.
The best score passed 85% in Oct 2019
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs dropped it from launch posts once the best score passed 85%
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 26 Jun 2025 | Gemma 3n E4B | 81.6 | BoolQ 0-shot (PT) |
| 15 Apr 2025 | BitNet b1.58 2B4T | 80.2 | BoolQ |
| 12 Mar 2025 | Gemma 3 27B | 82.4 | BoolQ 0-shot (PT) |
| 26 Feb 2025 | Phi-4 Mini | 81.2 | BoolQ (2-shot) |
| 20 Aug 2024 | Phi-3.5-MoE | 84.6 | BoolQ (2-shot) |
| 27 Jun 2024 | Gemma 2 27B | 84.8 | BoolQ (0-shot) |
| 22 Apr 2024 | phi-3-medium 14B | 86.6 | BoolQ (0-Shot) |
| 18 Apr 2024 | Llama 3-70B | 79 | BoolQ (0-shot) [pretrained] |
| 21 Feb 2024 | Gemma 7B | 83.2 | BoolQ (0-shot) |
| 12 Dec 2023 | Phi-2 | 83.3 | BoolQ, Table 2 |
| 10 May 2023 | PaLM 2-L | 90.9 | BoolQ, 1-shot |
25 system cards report it, from 4 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 16 Apr 2025 | BitNet b1.58 2B4T technical report | 80.2 | BoolQ (0-shot; Acc) (Table 1, Table 3) | Capabilities |
| 25 Mar 2025 | Gemma 3 4B technical report | 72.3 | BoolQ, PT, 0-shot (Table 9) | Capabilities |
| 25 Mar 2025 | Gemma 3 27B technical report | 82.4 | BoolQ, PT, 0-shot (Table 9) | Capabilities |
| 25 Mar 2025 | Gemma 3 1B technical report | 63.2 | BoolQ, PT, 0-shot (Table 9) | Capabilities |
| 25 Mar 2025 | Gemma 3 12B technical report | 78.8 | BoolQ, PT, 0-shot (Table 9) | Capabilities |
| 3 Mar 2025 | Phi-4 Mini technical report | 81.2 | BoolQ (2-Shot) (Table 7) | Capabilities |
| 30 Aug 2024 | phi-3-small 7.4B technical report | 84.8 | BoolQ (2-Shot) (Section 3 table) | Capabilities |
| 30 Aug 2024 | phi-3-mini 3.8B technical report | 77.2 | BoolQ (2-Shot) (Section 3 table) | Capabilities |
| 30 Aug 2024 | phi-3-medium 14B technical report | 86.5 | BoolQ (2-Shot) (Section 3 table) | Capabilities |
| 30 Aug 2024 | Phi-3.5-MoE technical report | 84.6 | BoolQ (2-shot) (Table 3) | Capabilities |
| 30 Aug 2024 | Phi-3.5-mini technical report | 78 | BoolQ (2-shot) (Table 3) | Capabilities |
| 31 Jul 2024 | Gemma 2 9B technical report | 84.2 | Boolq, PT, 0-shot (Table 13) | Capabilities |
| 31 Jul 2024 | Gemma 2 27B technical report | 84.8 | Boolq, PT, 0-shot (Table 13) | Capabilities |
| 23 Jul 2024 | Llama 3.1-8B model card | 75 | BoolQ, 0-shot, acc_char, pretrained (Base pretrained models table) | Capabilities |
| 23 Jul 2024 | Llama 3.1-70B model card | 79.4 | BoolQ, 0-shot, acc_char, pretrained (Base pretrained models table) | Capabilities |
| 23 Jul 2024 | Llama 3.1-405B model card | 80 | BoolQ, 0-shot, acc_char, pretrained (Base pretrained models table) | Capabilities |
| 18 Jun 2024 | ChatGLM3-6B technical report | 87.9 | ChatGLM3-6B-Base, BoolQ (Table 1) | Capabilities |
| 18 Apr 2024 | Llama 3-8B model card | 75.7 | BoolQ (0-shot), pretrained (Base pretrained models table) | Capabilities |
| 18 Apr 2024 | Llama 3-70B model card | 79 | BoolQ (0-shot), pretrained (Base pretrained models table) | Capabilities |
| 13 Mar 2024 | Gemma 7B technical report | 83.2 | Boolq, PT, 0-shot (Table 6) | Capabilities |
| 13 Mar 2024 | Gemma 2B technical report | 69.4 | Boolq, PT, 0-shot (Table 6) | Capabilities |
| 19 Jul 2023 | Llama 2-7B technical report | 77.4 | BoolQ, 0-shot, pretrained (Table 20) | Capabilities |
| 19 Jul 2023 | Llama 2-70B technical report | 85 | BoolQ, 0-shot, pretrained (Table 20) | Capabilities |
| 19 Jul 2023 | Llama 2-34B technical report | 83.7 | BoolQ, 0-shot, pretrained (Table 20) | Capabilities |
| 19 Jul 2023 | Llama 2-13B technical report | 81.7 | BoolQ, 0-shot, pretrained (Table 20) | Capabilities |
More in Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects.
Intern-S2-397B 89.8% - SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy.
Qwen 3.6 Max (Preview) 73.9% - WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1.
Tencent Hy4 preview 83.9% - KMMLU-Pro In use Models answer multiple-choice questions taken from Korean National Professional Licensure exams. It tests professional knowledge used in Korean industry.
Solar Pro 4 79.2% - BIG-Bench Extra Hard In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is a harmonic mean of task accuracies. Gemma 4 31B IT 74.4%
- HLE-Verified In use Verified and revised version of Humanity's Last Exam (1,811 items) with uncertain items removed.
Seed2.0 Pro 73.6%