Research map / Large language models / Reasoning and chain of thought
Search and self-correction in reasoning
Improve reasoning with trial-and-error, calibration, rollback, and selection-inference loops.
123 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Reasoning and chain of thought, keyword cluster “chain thought · thought cot · cot reasoning”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 4 tracked benchmarks.
33 papers in 2023–24 and 87 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 25.4% to 6.1%.
Open questions the data points to
- MATH (accuracy): best went from 73 (up to 2024) to 75.4 in 2025 — DOTS: Learning to Reason Dynamically in LLMs via Optimal Reasoning Trajectories Search (ICLR 2025 · 0 citations)
What looks settled
- GSM8K (accuracy): best up to 2024 89, best in 2025–26 89.7 from 2 papers, less than 1% better.
No longer reported by papers here: StrategyQA (accuracy, last in 2024); SVAMP (accuracy, last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Preferential Structures for Comparative Probabilistic Reasoning (AAAI 2017 · 5 citations)
- 2022 · Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (NeurIPS 2022 · 2,251 citations)
- 2023 · Tree of Thoughts: Deliberate Problem Solving with Large Language Models (NeurIPS 2023 · 566 citations)
- 2024 · Graph of Thoughts: Solving Elaborate Problems with Large Language Models (AAAI 2024 · 512 citations)
- 2025 · Is Sarcasm Detection a Step-by-Step Reasoning Process in Large Language Models? (AAAI 2025 · 16 citations)
- 2026 · NeSTR: A Neuro-Symbolic Abductive Framework for Temporal Reasoning in Large Language Models (AAAI 2026 · 3 citations)
Earliest papers here
- Preferential Structures for Comparative Probabilistic Reasoning (AAAI 2017 · 5 citations)
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (NeurIPS 2022 · 2,251 citations)
- STaR: Bootstrapping Reasoning With Reasoning (NeurIPS 2022 · 114 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
GSM8K · accuracy (higher is better)
9 papers, 2022–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 63.1 | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (NeurIPS 2022 · 2,251 citations) |
| 2023 | 82.9 | Complexity-Based Prompting for Multi-step Reasoning (ICLR 2023 · 73 citations) |
| 2024 | 89 | Algorithm of Thoughts: Enhancing Exploration of Ideas in Large Language Models (ICML 2024 · 13 citations) |
| 2025 | 89.7 | BRiTE: Bootstrapping Reinforced Thinking Process to Enhance Language Model Reasoning (ICML 2025 · 0 citations) |
MATH · accuracy (higher is better)
5 papers, 2024–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 73 | SELF-DISCOVER: Large Language Models Self-Compose Reasoning Structures (NeurIPS 2024 · 8 citations) |
| 2025 | 75.4 | DOTS: Learning to Reason Dynamically in LLMs via Optimal Reasoning Trajectories Search (ICLR 2025 · 0 citations) |
StrategyQA · accuracy (higher is better)
5 papers, 2023–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 77 | Complexity-Based Prompting for Multi-step Reasoning (ICLR 2023 · 73 citations) |
| 2024 | 84 | Algorithm of Thoughts: Enhancing Exploration of Ideas in Large Language Models (ICML 2024 · 13 citations) |
SVAMP · accuracy (higher is better)
4 papers, 2022–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 79 | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (NeurIPS 2022 · 2,251 citations) |
| 2024 | 93.7 | Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution (ICML 2024 · 26 citations) |
What has been tried
57 papers here report 653 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 7%, hurt 19%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use CoT instead of direct answering | MMLU-Pro · accuracy | 53.5 -> 72.6 | Helped | MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark (NeurIPS 2024 · 82 citations) |
| remove any self-referential operator | Maths datasets and ETHOS · accuracy | Hurt | Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution (ICML 2024 · 26 citations) | |
| Add linguistic cues to GoC | Sarcasm detection, average across four datasets · Macro-F1 | 72.78 -> 72.97 | Mixed | Is Sarcasm Detection a Step-by-Step Reasoning Process in Large Language Models? (AAAI 2025 · 16 citations) |
| Add CoT prompting | ASQA · Fluency | 76.3 -> 82.0 | Helped | Chain-of-Thought Improves Text Generation with Citations in Large Language Models (AAAI 2024 · 15 citations) |
| manually resolve the final answer after AoT finds a solution | Game of 24 · success | 71% -> 78% | Helped | Algorithm of Thoughts: Enhancing Exploration of Ideas in Large Language Models (ICML 2024 · 13 citations) |
| Use multi-stage step checking instead of global solution checking | MathQA · verification accuracy | 55.0% -> 66.7% | Helped | SelfCheck: Using LLMs to Zero-Shot Check Their Own Step-by-Step Reasoning (ICLR 2024 · 12 citations) |
| Use CoC (Interweave) rather than Python-only execution | BIG-Bench Hard, single task · accuracy (%) | 48 -> 84 | Helped | Chain of Code: Reasoning with a Language Model-Augmented Code Emulator (ICML 2024 · 12 citations) |
| Use subject-centric evidence chains without evidence chain summarization | bAbI task-1 · prediction accuracy | 0.607 -> 0.789 | Helped | MindMap: Constructing Evidence Chains for Multi-Step Reasoning in Large Language Models (AAAI 2024 · 8 citations) |
| remove negative data (dual LoRA on positive only) | MATH test set · average accuracy | 6.58 vs 6.81 | Hurt | Turning Dust into Gold: Distilling Complex Reasoning Capabilities from LLMs by Leveraging Negative Data (AAAI 2024 · 8 citations) |
| Add ADAPT to SELECT | 4 reasoning tasks · zero-shot reasoning performance | Helped | SELF-DISCOVER: Large Language Models Self-Compose Reasoning Structures (NeurIPS 2024 · 8 citations) | |
| Add clean example during CD-CoT step 3 | NoRa Base-9 irrelevant (medium) · accuracy | 56.7 -> 60.7 | Helped | Can Language Models Perform Robust Reasoning in Chain-of-thought Prompting with Noisy Rationales? (NeurIPS 2024 · 6 citations) |
| Add SFT to the GRPO-based training pipeline | Text2CAD test set · Mean CD (×10^3) | 76.20 -> 6.54 | Helped | CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward (NeurIPS 2025 · 4 citations) |
What reviewers will expect
Datasets and baselines named in the 87 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- GSM8K · 24 of 87 papers
- MATH · 15 of 87 papers
- MATH-500 · 9 of 87 papers
- MBPP · 7 of 87 papers
- HotpotQA · 6 of 87 papers
- AIME 2024 · 6 of 87 papers
- CommonsenseQA · 6 of 87 papers
- HumanEval · 6 of 87 papers
Baselines
- CoT · 7 of 87 papers
- SFT · 6 of 87 papers
- Qwen2.5-7B-Instruct · 4 of 87 papers
- DeepSeek-V3 · 4 of 87 papers
- Self-Refine · 4 of 87 papers
- Coconut · 4 of 87 papers
- Self-Consistency · 4 of 87 papers
- QwQ-32B · 4 of 87 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 120 papers; from OpenAlex and the papers' affiliations).
- Google DeepMind · 7 papers
- Google · 7 papers
- Stanford University · 4 papers
- Microsoft · 4 papers
- Tsinghua University · 3 papers
- Peking University · 3 papers
- Amazon · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- Alibaba Group · 3 papers
- Wuhan University · 2 papers
Use it
Download BibTeX of 31 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Chain-of-thought training and analysis (127 papers)
- Long CoT dynamics and efficiency (136 papers)
- Theory of autoregressive reasoning (28 papers)
Elsewhere in the topic
- RL with verifiable rewards for reasoning (270 papers)
- Efficient and adaptive thinking (231 papers)
- Test-time compute scaling (205 papers)
- Routing, pruning and distillation for reasoning (135 papers)
- Efficient LLM serving (126 papers)
- Speculative decoding (92 papers)
- KV cache compression (45 papers)
- Process reward models (39 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).