LLM Uncertainty Bench
Models answer questions across five language tasks. The original benchmark measures accuracy and uncertainty, while the port checks only the selected answer.
In October 2024, GPT-4 led the paper’s MMLU sampling comparison at 81.75% accuracy. The authors also tested GPT-3.5 and Qwen-72B.
- Released
- January 2024
- Kind
- Language reasoning
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
-
Llama 2-70B #210 on the capability index
- Last model added
- November 2023
- Scores refreshed
- November 20th, 2026
Yi-34B leads with 81.5% among models released by November 2023 Yi-34B leads with 81.5% among models released by November 2023
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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 November 2023, 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
- 1
Yi-34B 81.5%
- 2 Qwen-72B 78%
- 3
Qwen-14B 74%
- 4
Llama 2-70B 72.2%
- 5
DeepSeek LLM 67B 71.7%
- 6
Yi 6B 69%
- 7
Mistral 7B v0.1 64.1%
- 8
Llama 2-13B 60.4%
- 9
Qwen-7B 59.9%
- 10 internlm-7b 49.3%
- 11
Llama 2-7B 46.5%
- 12 DeepSeek-7B 45.9%
- 13 Qwen-1_8B 42.3%
- 14
Falcon-40B 34.5%
- 15
MPT-7B 27.2%
- 16
Falcon-7B 24.8%
Source: Official LLM Uncertainty Bench leaderboard
- 1
Yi-34B 81.5%
- 2
Qwen-14B 74%
- 3
Llama 2-70B 72.2%
- 4
DeepSeek LLM 67B 71.7%
- 5
Yi 6B 69%
- 6
Mistral 7B v0.1 64.1%
- 7
Llama 2-13B 60.4%
- 8
Qwen-7B 59.9%
- 9 internlm-7b 49.3%
- 10
Llama 2-7B 46.5%
- 11 Qwen-1_8B 42.3%
- 12
Falcon-40B 34.5%
- 13
MPT-7B 27.2%
- 14
Falcon-7B 24.8%
Source: Official LLM Uncertainty Bench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 81.5% | Nov 2023 | Official leaderboard | |
| 2 | Qwen-72B | 78% | Official leaderboard | ||
| 3 | | 74% | Sep 2023 | Official leaderboard | |
| 4 | | 72.2% | Jul 2023 | Official leaderboard | |
| 5 | | 71.7% | Nov 2023 | Official leaderboard | |
| 6 | | 69% | Nov 2023 | Official leaderboard | |
| 7 | | 64.1% | Sep 2023 | Official leaderboard | |
| 8 | | 60.4% | Jul 2023 | Official leaderboard | |
| 9 | | 59.9% | Sep 2023 | Official leaderboard | |
| 10 | internlm-7b | 49.3% | Jul 2023 | Official leaderboard | |
| 11 | | 46.5% | Jul 2023 | Official leaderboard | |
| 12 | DeepSeek-7B | 45.9% | Official leaderboard | ||
| 13 | Qwen-1_8B | 42.3% | Nov 2023 | Official leaderboard | |
| 14 | | 34.5% | Mar 2023 | Official leaderboard | |
| 15 | | 27.2% | May 2023 | Official leaderboard | |
| 16 | | 24.8% | Apr 2023 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 81.5% | Nov 2023 | Official leaderboard | |
| 2 | | 74% | Sep 2023 | Official leaderboard | |
| 3 | | 72.2% | Jul 2023 | Official leaderboard | |
| 4 | | 71.7% | Nov 2023 | Official leaderboard | |
| 5 | | 69% | Nov 2023 | Official leaderboard | |
| 6 | | 64.1% | Sep 2023 | Official leaderboard | |
| 7 | | 60.4% | Jul 2023 | Official leaderboard | |
| 8 | | 59.9% | Sep 2023 | Official leaderboard | |
| 9 | internlm-7b | 49.3% | Jul 2023 | Official leaderboard | |
| 10 | | 46.5% | Jul 2023 | Official leaderboard | |
| 11 | Qwen-1_8B | 42.3% | Nov 2023 | Official leaderboard | |
| 12 | | 34.5% | Mar 2023 | Official leaderboard | |
| 13 | | 27.2% | May 2023 | Official leaderboard | |
| 14 | | 24.8% | Apr 2023 | Official leaderboard |
Rankings by part
We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Question Answering, Reading Comprehension and Commonsense Inference. Yi-34B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Question Answering
14 models we track have a score here.
- 1
Yi-34B 71.2%
- 2
DeepSeek LLM 67B ↑2 68%
- 3
Llama 2-70B 65.9%
- 4
Qwen-14B ↓2 64.3%
- 5
Mistral 7B v0.1 ↑1 60.4%
- 6
Yi 6B ↓1 57.6%
- 7
Qwen-7B ↑1 55.2%
- 8
Llama 2-13B ↓1 52.5%
- 9 internlm-7b 48.4%
- 10
Llama 2-7B 45.6%
Reading Comprehension
14 models we track have a score here.
- 1
Yi-34B 94.5%
- 2
Qwen-14B 91.5%
- 3
Llama 2-70B 89.3%
- 4
DeepSeek LLM 67B 88.7%
- 5
Yi 6B 86%
- 6
Qwen-7B ↑2 83.9%
- 7
Mistral 7B v0.1 ↓1 81.9%
- 8
Llama 2-13B ↓1 77.2%
- 9 internlm-7b 73.9%
- 10
Llama 2-7B 65.8%
Commonsense Inference
14 models we track have a score here.
- 1
Yi-34B 94%
- 2
Qwen-14B 91%
- 3
Llama 2-70B 81.6%
- 4
Yi 6B ↑1 76.5%
- 5
DeepSeek LLM 67B ↓1 70.8%
- 6
Qwen-7B ↑2 63.7%
- 7
Mistral 7B v0.1 ↓1 62.9%
- 8
Llama 2-13B ↓1 59.7%
- 9 internlm-7b 46.2%
- 10
Llama 2-7B 43%
Dialogue Response Selection
14 models we track have a score here.
- 1
Yi-34B 76.1%
- 2
DeepSeek LLM 67B ↑2 74.7%
- 3
Qwen-14B ↓1 73.9%
- 4
Llama 2-70B ↓1 67%
- 5
Qwen-7B ↑3 64%
- 6
Yi 6B ↓1 58.7%
- 7
Mistral 7B v0.1 ↓1 53.2%
- 8
Llama 2-13B ↓1 52.6%
- 9 internlm-7b 43.7%
- 10 Qwen-1_8B ↑1 35.5%
Document Summarization
14 models we track have a score here.
- 1
Yi-34B 71.5%
- 2
Yi 6B ↑3 66.1%
- 3
Mistral 7B v0.1 ↑3 62.2%
- 4
Llama 2-13B ↑3 60%
- 5
Llama 2-70B ↓2 57.4%
- 6
DeepSeek LLM 67B ↓2 56.2%
- 7
Qwen-14B ↓5 49.3%
- 8
Llama 2-7B ↑2 45.6%
- 9 internlm-7b 34.4%
- 10
Qwen-7B ↓2 32.5%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official LLM Uncertainty Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official LLM Uncertainty Bench 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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 3 years ago, in November 2023, for Qwen-1_8B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (16 entries).
- Pass Models scored 14 models we track, plus 2 other entries on its leaderboard.
- Warning Newest score The newest came 3 years ago, in November 2023, for Qwen-1_8B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Llama 2-70B (#210).
- Pass Headroom The best is 81.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 14 rows for models we track and 2 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.