SATBench
Models decide whether a set of constraints in a narrative puzzle can all hold at once. A Boolean satisfiability solver supplies the correct label.
The Harbor comparison available November 19, 2026 reports GPT-5 nano alone on 100 sampled puzzles. It scored 72.0% originally and 70.3% on Harbor.
- Made by
- Anjiang Wei and colleagues
- Released
- May 2025
- Kind
- Language reasoning
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- o4-mini #95 on the capability index
- Last model added
- April 2025
- Scores refreshed
- November 20th, 2026
o4-mini leads with 89.3% among models released by April 2025 o4-mini leads with 89.3% among models released by April 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 89.3 o4-mini
- 87.8 DeepSeek-R1 (snapshot unspecified)
- 84.0 DeepSeek-V3 (snapshot unspecified)
- 74.8 Claude 3.7 Sonnet
- 68.3
QwQ-32B
- 68.1 Qwen3-235B-Int8
- 66.4
Qwen3-14B
- 66.2
DeepSeek-R1-Distill-Qwen-32B
- 62.6
Qwen3-8B
- 62.6
Qwen3-4B
- 62.1
Llama 4 Maverick
- 62.0
DeepSeek-R1-Distill-Qwen-14B
- 58.7
Llama 3.3 70B
- 58.0 gpt-4o (snapshot unspecified)
- 57.7
Llama 3.1-70B
- 56.9
Llama 4 Scout
- 53.9 gpt-4o-mini (snapshot unspecified)
- 53.7
Qwen3-1.7B
- 43.9
DeepSeek-R1-Distill-Qwen-7B
- 38.2
Llama 3.1-8B
- 1 o4-mini Zero-shot 89.3%
- 2 DeepSeek-R1 (snapshot unspecified) Zero-shot 87.8%
- 3 DeepSeek-V3 (snapshot unspecified) Zero-shot 84%
- 4 Claude 3.7 Sonnet Zero-shot 74.8%
- 5
QwQ-32B Zero-shot 68.3%
- 6 Qwen3-235B-Int8 Zero-shot 68.1%
- 7
Qwen3-14B Zero-shot 66.4%
- 8
DeepSeek-R1-Distill-Qwen-32B Zero-shot 66.2%
- 9
Qwen3-8B Zero-shot 62.6%
- 10
Qwen3-4B Zero-shot 62.6%
- 11
Llama 4 Maverick Zero-shot 62.1%
- 12
DeepSeek-R1-Distill-Qwen-14B Zero-shot 62%
- 13
Llama 3.3 70B Zero-shot 58.7%
- 14 gpt-4o (snapshot unspecified) Zero-shot 58%
- 15
Llama 3.1-70B Zero-shot 57.7%
- 16
Llama 4 Scout Zero-shot 56.9%
- 17 gpt-4o-mini (snapshot unspecified) Zero-shot 53.9%
- 18
Qwen3-1.7B Zero-shot 53.7%
- 19
DeepSeek-R1-Distill-Qwen-7B Zero-shot 43.9%
- 20
Llama 3.1-8B Zero-shot 38.2%
Source: The SATBench paper
- 89.3 o4-mini
- 74.8 Claude 3.7 Sonnet
- 68.3
QwQ-32B
- 66.4
Qwen3-14B
- 66.2
DeepSeek-R1-Distill-Qwen-32B
- 62.6
Qwen3-8B
- 62.6
Qwen3-4B
- 62.1
Llama 4 Maverick
- 62.0
DeepSeek-R1-Distill-Qwen-14B
- 58.7
Llama 3.3 70B
- 57.7
Llama 3.1-70B
- 56.9
Llama 4 Scout
- 53.7
Qwen3-1.7B
- 43.9
DeepSeek-R1-Distill-Qwen-7B
- 38.2
Llama 3.1-8B
- 1 o4-mini Zero-shot 89.3%
- 2 Claude 3.7 Sonnet Zero-shot 74.8%
- 3
QwQ-32B Zero-shot 68.3%
- 4
Qwen3-14B Zero-shot 66.4%
- 5
DeepSeek-R1-Distill-Qwen-32B Zero-shot 66.2%
- 6
Qwen3-8B Zero-shot 62.6%
- 7
Qwen3-4B Zero-shot 62.6%
- 8
Llama 4 Maverick Zero-shot 62.1%
- 9
DeepSeek-R1-Distill-Qwen-14B Zero-shot 62%
- 10
Llama 3.3 70B Zero-shot 58.7%
- 11
Llama 3.1-70B Zero-shot 57.7%
- 12
Llama 4 Scout Zero-shot 56.9%
- 13
Qwen3-1.7B Zero-shot 53.7%
- 14
DeepSeek-R1-Distill-Qwen-7B Zero-shot 43.9%
- 15
Llama 3.1-8B Zero-shot 38.2%
Source: The SATBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o4-mini | 89.3% | Zero-shot | Apr 2025 | Official leaderboard |
| 2 | DeepSeek-R1 (snapshot unspecified) | 87.8% | Zero-shot | Official leaderboard | |
| 3 | DeepSeek-V3 (snapshot unspecified) | 84% | Zero-shot | Official leaderboard | |
| 4 | Claude 3.7 Sonnet | 74.8% | Zero-shot | Feb 2025 | Official leaderboard |
| 5 | | 68.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 6 | Qwen3-235B-Int8 | 68.1% | Zero-shot | Official leaderboard | |
| 7 | | 66.4% | Zero-shot | Apr 2025 | Official leaderboard |
| 8 | | 66.2% | Zero-shot | Jan 2025 | Official leaderboard |
| 9 | | 62.6% | Zero-shot | Apr 2025 | Official leaderboard |
| 10 | | 62.6% | Zero-shot | Apr 2025 | Official leaderboard |
| 11 | | 62.1% | Zero-shot | Apr 2025 | Official leaderboard |
| 12 | | 62% | Zero-shot | Jan 2025 | Official leaderboard |
| 13 | | 58.7% | Zero-shot | Dec 2024 | Official leaderboard |
| 14 | gpt-4o (snapshot unspecified) | 58% | Zero-shot | Official leaderboard | |
| 15 | | 57.7% | Zero-shot | Jul 2024 | Official leaderboard |
| 16 | | 56.9% | Zero-shot | Apr 2025 | Official leaderboard |
| 17 | gpt-4o-mini (snapshot unspecified) | 53.9% | Zero-shot | Official leaderboard | |
| 18 | | 53.7% | Zero-shot | Apr 2025 | Official leaderboard |
| 19 | | 43.9% | Zero-shot | Jan 2025 | Official leaderboard |
| 20 | | 38.2% | Zero-shot | Jul 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o4-mini | 89.3% | Zero-shot | Apr 2025 | Official leaderboard |
| 2 | Claude 3.7 Sonnet | 74.8% | Zero-shot | Feb 2025 | Official leaderboard |
| 3 | | 68.3% | Zero-shot | Mar 2025 | Official leaderboard |
| 4 | | 66.4% | Zero-shot | Apr 2025 | Official leaderboard |
| 5 | | 66.2% | Zero-shot | Jan 2025 | Official leaderboard |
| 6 | | 62.6% | Zero-shot | Apr 2025 | Official leaderboard |
| 7 | | 62.6% | Zero-shot | Apr 2025 | Official leaderboard |
| 8 | | 62.1% | Zero-shot | Apr 2025 | Official leaderboard |
| 9 | | 62% | Zero-shot | Jan 2025 | Official leaderboard |
| 10 | | 58.7% | Zero-shot | Dec 2024 | Official leaderboard |
| 11 | | 57.7% | Zero-shot | Jul 2024 | Official leaderboard |
| 12 | | 56.9% | Zero-shot | Apr 2025 | Official leaderboard |
| 13 | | 53.7% | Zero-shot | Apr 2025 | Official leaderboard |
| 14 | | 43.9% | Zero-shot | Jan 2025 | Official leaderboard |
| 15 | | 38.2% | Zero-shot | Jul 2024 | Official leaderboard |
Rankings by part
We read 9 parts from the benchmark’s paper in 3 groups: sat, unsat and overall. Each ranks models on one part of the test. o4-mini leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Easy
15 models we track have a score here.
- 1 o4-mini 97%
- 2
QwQ-32B ↑1 92.5%
- 3
Llama 3.3 70B ↑7 90.7%
- 4 Claude 3.7 Sonnet ↓2 88.4%
- 5
Qwen3-14B ↓1 87.1%
- 6
DeepSeek-R1-Distill-Qwen-32B ↓1 84.5%
- 7
Llama 4 Scout ↑5 84.3%
- 8
Qwen3-4B ↓1 84.1%
- 9
DeepSeek-R1-Distill-Qwen-14B 82.9%
- 10
Qwen3-8B ↓4 82.7%
Medium
15 models we track have a score here.
- 1 o4-mini 96.7%
- 2
Llama 3.3 70B ↑8 89%
- 3
Llama 4 Maverick ↑5 86.2%
- 4
Qwen3-4B ↑3 78.1%
- 5 Claude 3.7 Sonnet ↓3 77.6%
- 6
Llama 4 Scout ↑6 76.7%
- 7
Qwen3-8B ↓1 76.7%
- 8
QwQ-32B ↓5 75.7%
- 9
Qwen3-14B ↓5 72.9%
- 10
Qwen3-1.7B ↑3 65.7%
Hard
15 models we track have a score here.
- 1 o4-mini 91.1%
- 2
Llama 4 Maverick ↑6 86.1%
- 3 Claude 3.7 Sonnet ↓1 83.6%
- 4
Qwen3-14B 80%
- 5
Qwen3-4B ↑2 78.6%
- 6
Llama 3.3 70B ↑4 75.7%
- 7
Qwen3-8B ↓1 67.5%
- 8
Llama 4 Scout ↑4 66.4%
- 9
QwQ-32B ↓6 59.3%
- 10
Qwen3-1.7B ↑3 53.2%
Easy
15 models we track have a score here.
- 1 o4-mini 98.2%
- 2 Claude 3.7 Sonnet 93.8%
- 3
DeepSeek-R1-Distill-Qwen-32B ↑2 90%
- 4
Qwen3-14B 88.9%
- 5
DeepSeek-R1-Distill-Qwen-14B ↑4 85.7%
- 6
QwQ-32B ↓3 84.1%
- 7
Qwen3-8B ↓1 81.6%
- 8
Qwen3-4B ↓1 80.7%
- 9
Llama 4 Maverick ↓1 76.8%
- 10
DeepSeek-R1-Distill-Qwen-7B ↑4 69.1%
Medium
15 models we track have a score here.
- 1 o4-mini 88.1%
- 2
DeepSeek-R1-Distill-Qwen-32B ↑3 68.1%
- 3 Claude 3.7 Sonnet ↓1 63.3%
- 4
Llama 3.1-70B ↑7 59%
- 5
DeepSeek-R1-Distill-Qwen-14B ↑4 59%
- 6
QwQ-32B ↓3 51.9%
- 7
Qwen3-14B ↓3 47.6%
- 8
DeepSeek-R1-Distill-Qwen-7B ↑6 43.8%
- 9
Qwen3-8B ↓3 34.8%
- 10
Qwen3-4B ↓3 31.9%
Hard
15 models we track have a score here.
- 1 o4-mini 65%
- 2
DeepSeek-R1-Distill-Qwen-32B ↑3 58.6%
- 3
DeepSeek-R1-Distill-Qwen-14B ↑6 51.8%
- 4
Llama 3.1-70B ↑7 48.9%
- 5
QwQ-32B ↓2 46.4%
- 6
Qwen3-1.7B ↑7 42.5%
- 7
DeepSeek-R1-Distill-Qwen-7B ↑7 42.1%
- 8 Claude 3.7 Sonnet ↓6 42.1%
- 9
Llama 4 Scout ↑3 37.5%
- 10
Qwen3-8B ↓4 32.1%
Easy
15 models we track have a score here.
- 1 o4-mini 97.6%
- 2 Claude 3.7 Sonnet 91.1%
- 3
QwQ-32B 88.3%
- 4
Qwen3-14B 88%
- 5
DeepSeek-R1-Distill-Qwen-32B 87.2%
- 6
DeepSeek-R1-Distill-Qwen-14B ↑3 84.3%
- 7
Qwen3-4B 82.4%
- 8
Qwen3-8B ↓2 82.1%
- 9
Llama 4 Maverick ↓1 78.5%
- 10
Llama 3.1-70B ↑1 68.6%
Medium
15 models we track have a score here.
- 1 o4-mini 92.4%
- 2 Claude 3.7 Sonnet 70.5%
- 3
QwQ-32B 63.8%
- 4
DeepSeek-R1-Distill-Qwen-32B ↑1 61%
- 5
Qwen3-14B ↓1 60.2%
- 6
Llama 3.3 70B ↑4 58.1%
- 7
Llama 3.1-70B ↑4 57.4%
- 8
Llama 4 Maverick 56%
- 9
Qwen3-8B ↓3 55.7%
- 10
DeepSeek-R1-Distill-Qwen-14B ↓1 55.2%
Hard
15 models we track have a score here.
- 1 o4-mini 78%
- 2 Claude 3.7 Sonnet 62.9%
- 3
Llama 3.3 70B ↑7 52.9%
- 4
QwQ-32B ↓1 52.9%
- 5
Llama 4 Maverick ↑3 52%
- 6
Llama 4 Scout ↑6 52%
- 7
Qwen3-14B ↓3 51.1%
- 8
DeepSeek-R1-Distill-Qwen-32B ↓3 50.4%
- 9
Qwen3-4B ↓2 50.4%
- 10
Qwen3-8B ↓4 49.8%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The SATBench paper
O4-mini on the official leaderboard passed 85% in April 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official SATBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
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%
- MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy.
Seed1.8 92.3%
- 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%
- BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly.
Gemma 4 31B IT 74.4%
- HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (20 entries).
- Pass Models scored 15 models we track, plus 5 other entries on its leaderboard.
- Note Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. That’s expected for a saturated test.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is o4-mini (#95).
- Note Headroom Saturated. The best is 89.3%, past 85% since April 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on aclanthology.org: 15 rows for models we track and 5 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.