Benchmark Almanac

Corr2Cause

RoBERTa-Large MNLI 94.7% tops its leaderboard Llama 2-7B 92.9% leads the models we track
Not in model cards

Models decide whether a causal claim follows from statements about correlations and independences among variables. The original study reports F1 for detecting valid claims.

The makers’ repository has 115 stars. The CaDO team used the official development and test sets.

Made by
CausalNLP and collaborators
Released
May 2023
Kind
Language reasoning
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Subject
Mathematics
Language
English
In and out
Text → Text
Interaction
Single turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Llama 2-7B #238 on the capability index
Last model added
July 2023
Scores refreshed
November 20th, 2026
29
Entries in its paper, 5 models we track

RoBERTa-Large MNLI tops the leaderboard with 94.7% Llama 2-7B leads with 92.9% among models released by July 2023

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. The best model we track is Llama 2-7B, at 92.9%. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2023, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 RoBERTa-Large MNLI Fine-tuned 94.7%
  2. 2 GPT2-XL Fine-tuned 94.3%
  3. 3 GPT2-Large Fine-tuned 94.3%
  4. 4 RoBERTa-Base MNLI Fine-tuned 94.3%
  5. 5 Llama 2-7B Fine-tuned 92.9%
  6. 6 LLaMA-7B Fine-tuned 92%
  7. 7 BERT-Large MNLI Fine-tuned 90.2%
  8. 8 BERT-Base MNLI Fine-tuned 89.9%
  9. 9 GPT2 Fine-tuned 89.2%
  10. 10 RoBERTa-Large Fine-tuned 89.1%
  11. 11 RoBERTa-Base Fine-tuned 87.6%
  12. 12 GPT-3 175B (davinci) Fine-tuned 85.5%
  13. 13 BERT-Large Fine-tuned 85.3%
  14. 14 GPT-3 Curie Fine-tuned 81.2%
  15. 15 GPT-3 Ada Fine-tuned 79.8%
  16. 16 GPT-3 Babbage Fine-tuned 78.2%
  17. 17 BERT-Base Fine-tuned 69.3%
  18. 18 BART MNLI Original query 33.4%
  19. 19 GPT-4 (snapshot unspecified) Original query 29.1%
  20. 20 Alpaca-7B Original query 27.4%
  21. 21 DistilBART MNLI Original query 26.7%
  22. 22 GPT-3.5 (snapshot unspecified) Enhanced query 25.4%
  23. 23 RoBERTa MNLI Original query 22.8%
  24. 24 davinci-002 Original query 21.9%
  25. 25 DistilBERT MNLI Original query 20.7%
  26. 26 GPT-3 Instruct (text-davinci-001) Original query 18%
  27. 27 davinci-003 Original query 15.7%
  28. 28 DeBERTa MNLI Original query 14.5%
  29. 29 BERT MNLI Original query 2.82%

Source: The Corr2Cause paper

#EntryScoreSettingDateSource
1 RoBERTa-Large MNLI 94.7% Fine-tuned Official leaderboard
2 GPT2-XL 94.3% Fine-tuned Official leaderboard
3 GPT2-Large 94.3% Fine-tuned Official leaderboard
4 RoBERTa-Base MNLI 94.3% Fine-tuned Official leaderboard
5 Llama 2-7B 92.9% Fine-tuned Jul 2023 Official leaderboard
6 LLaMA-7B 92% Fine-tuned Feb 2023 Official leaderboard
7 BERT-Large MNLI 90.2% Fine-tuned Official leaderboard
8 BERT-Base MNLI 89.9% Fine-tuned Official leaderboard
9 GPT2 89.2% Fine-tuned Official leaderboard
10 RoBERTa-Large 89.1% Fine-tuned Official leaderboard
11 RoBERTa-Base 87.6% Fine-tuned Official leaderboard
12 GPT-3 175B (davinci) 85.5% Fine-tuned May 2020 Official leaderboard
13 BERT-Large 85.3% Fine-tuned Official leaderboard
14 GPT-3 Curie 81.2% Fine-tuned Official leaderboard
15 GPT-3 Ada 79.8% Fine-tuned Official leaderboard
16 GPT-3 Babbage 78.2% Fine-tuned Official leaderboard
17 BERT-Base 69.3% Fine-tuned Official leaderboard
18 BART MNLI 33.4% Original query Official leaderboard
19 GPT-4 (snapshot unspecified) 29.1% Original query Official leaderboard
20 Alpaca-7B 27.4% Original query Official leaderboard
21 DistilBART MNLI 26.7% Original query Official leaderboard
22 GPT-3.5 (snapshot unspecified) 25.4% Enhanced query Official leaderboard
23 RoBERTa MNLI 22.8% Original query Official leaderboard
24 davinci-002 21.9% Original query Mar 2022 Official leaderboard
25 DistilBERT MNLI 20.7% Original query Official leaderboard
26 GPT-3 Instruct (text-davinci-001) 18% Original query Official leaderboard
27 davinci-003 15.7% Original query Nov 2022 Official leaderboard
28 DeBERTa MNLI 14.5% Original query Official leaderboard
29 BERT MNLI 2.82% Original query Official leaderboard
#ModelScoreSettingDateSource
1 Llama 2-7B 92.9% Fine-tuned Jul 2023 Official leaderboard
2 LLaMA-7B 92% Fine-tuned Feb 2023 Official leaderboard
3 GPT-3 175B (davinci) 85.5% Fine-tuned May 2020 Official leaderboard
4 davinci-002 21.9% Original query Mar 2022 Official leaderboard
5 davinci-003 15.7% Original query Nov 2022 Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s paper: Paraphrase and Variable refactorization. Each ranks models on one part of the test. GPT-3 175B (davinci) leads Paraphrase and LLaMA-7B leads Variable refactorization. Arrows show how far a model moved from its overall rank among the models we track.

Paraphrase

3 models we track have a score here.

  1. 1 GPT-3 175B (davinci) ↑2 65%
  2. 2 LLaMA-7B 56.4%
  3. 3 Llama 2-7B ↓2 52.2%

2 models have held number one in the benchmark’s paper since its release in May 2023

Lab claims count from the day the card came out, independent runs from the model’s release.

Meta
Benchmark paper Passed 85% in May 2023, by LLaMA-7B
Llama 2-7B 92.9%

Source: The Corr2Cause paper

GPT-3 175B (davinci) on the official leaderboard passed 85% in May 2020

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.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2020 2021 2022 2023 Near the ceiling (85%) Llama 2-7B GPT-3 175B (davinci) 0% 20% 40% 60% 80% 100% 2020 2021 2022 2023 Near the ceiling (85%) Llama 2-7B GPT-3 175B (davinci)

Source: Official Corr2Cause 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

All 332 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 (29 entries).
  • Pass Models scored 5 models we track, plus 24 other entries on its leaderboard.
  • Note Newest score The newest came 3 years ago, in July 2023, for Llama 2-7B. 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 Llama 2-7B (#238).
  • Note Headroom Saturated. The best is 92.9%, past 85% since May 2020.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 7 rows for models we track and 24 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.

How we check