Research map / Large language models / Reasoning and chain of thought
Test-time compute scaling
Allocate inference-time compute through confidence, branching, and self-consistency strategies.
205 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Reasoning and chain of thought, keyword cluster “test time · scaling · rl”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
4 papers in 2023–24 and 200 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.1% to 14.1%.
Open questions the data points to
- Papers report CoT on GSM8k (accuracy) anywhere from 42.9 to 88.1: comparisons against it depend on whose number you take.
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
- 2021 · To Beam Or Not To Beam: That is a Question of Cooperation for Language GANs (NeurIPS 2021 · 1 citation)
- 2023 · Inference-Time Intervention: Eliciting Truthful Answers from a Language Model (NeurIPS 2023 · 56 citations)
- 2024 · ConsistentEE: A Consistent and Hardness-Guided Early Exiting Method for Accelerating Language Models Inference (AAAI 2024 · 9 citations)
- 2025 · Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models (ICLR 2025 · 5 citations)
- 2026 · The Illusion of Diminishing Returns: Measuring Long Horizon Execution in LLMs (ICLR 2026 · 1 citation)
Earliest papers here
- To Beam Or Not To Beam: That is a Question of Cooperation for Language GANs (NeurIPS 2021 · 1 citation)
- Inference-Time Intervention: Eliciting Truthful Answers from a Language Model (NeurIPS 2023 · 56 citations)
- ConsistentEE: A Consistent and Hardness-Guided Early Exiting Method for Accelerating Language Models Inference (AAAI 2024 · 9 citations)
What has been tried
46 papers here report 536 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 7%, hurt 18%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Hardness-guided reward (memorized layer) vs vanilla reward | ablation under varied alpha (GLUE) · accuracy vs speedup trade-off | Helped | ConsistentEE: A Consistent and Hardness-Guided Early Exiting Method for Accelerating Language Models Inference (AAAI 2024 · 9 citations) | |
| Lower sampling temperature from 0.7 to 0.5 | GSM8K · gap | Hurt | Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models (ICLR 2025 · 5 citations) | |
| Use three principal components to represent benchmark capabilities | Collected base-model benchmarks · explained variance | Helped | Observational Scaling Laws and the Predictability of Langauge Model Performance (NeurIPS 2024 · 5 citations) | |
| Add KL regularization to EM-FT with N=1 | Math average · score | 40.2 -> 39.6 | Hurt | The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning (NeurIPS 2025 · 3 citations) |
| Add pause tokens during finetuning only, without pause-pretraining | GSM8k · accuracy | 7.5 ± 0.5 -> 6.9 ± 1.0 | Hurt | Think before you speak: Training Language Models With Pause Tokens (ICLR 2024 · 2 citations) |
| Use stepwise rather than full-trajectory preference data | StrongReject · goodness score | 0.6663 -> 0.6955 | Helped | STAIR: Improving Safety Alignment with Introspective Reasoning (ICML 2025 · 1 citation) |
| Remove Insight 1 and use random initialization | 2WikiMultiHopQA · Finding optimal configurations | Hurt | AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Tasks (NeurIPS 2025 · 1 citation) | |
| Add Reasoning Pruning to Perplexity Consistency | MathOdyssey · # samplings to reach self-consistency performance | 112 -> 32 | Helped | A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning (NeurIPS 2025 · 1 citation) |
| Add zero-shot chain-of-thought prompting for pairwise comparisons | GPQA and other evaluated settings · accuracy | Helped | Provable Scaling Laws for the Test-Time Compute of Large Language Models (NeurIPS 2025 · 1 citation) | |
| increase alpha_base from 0.1 to 1.0 | Macro Average · Pass@1 | 48.0 -> 39.2 | Hurt | The Overthinker's DIET: Cutting Token Calories with DIfficulty-AwarE Training (NeurIPS 2025 · 1 citation) |
| Use lower sampling temperature | APPS · Pass@token | Helped | DISC: Dynamic Decomposition Improves LLM Inference Scaling (NeurIPS 2025 · 1 citation) | |
| Vary the prior hyperparameters | LiveCodeBench · Pass@1 | 38.4 ± 1.6; 37.3 ± 0.4; 36.8 ± 1.5; 37.5 ± 1.5; 37.7 ± 1.3 | No clear effect | Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Branching Tree Search (NeurIPS 2025 · 1 citation) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| CoT | GSM8k | accuracy | 42.9 – 88.1 | 2 papers, e.g. NeurIPS 2025 and ICML 2025 |
What reviewers will expect
Datasets and baselines named in the 200 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 · 39 of 200 papers
- MATH-500 · 26 of 200 papers
- MATH · 25 of 200 papers
- MBPP · 17 of 200 papers
- AIME 2024 · 13 of 200 papers
- HumanEval · 12 of 200 papers
- LiveCodeBench · 9 of 200 papers
- AIME24 · 9 of 200 papers
Baselines
- Best-of-N · 17 of 200 papers
- GPT-4o · 14 of 200 papers
- GRPO · 14 of 200 papers
- Self-Consistency · 13 of 200 papers
- SFT · 10 of 200 papers
- Majority Voting · 8 of 200 papers
- Beam Search · 7 of 200 papers
- DeepSeek-R1 · 6 of 200 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 204 papers; from OpenAlex and the papers' affiliations).
- Google · 10 papers
- Alibaba Group · 7 papers
- Meta · 6 papers
- Tsinghua University · 6 papers
- Microsoft · 5 papers
- Stanford University · 5 papers
- Carnegie Mellon University · 4 papers
- Zhejiang University · 4 papers
- Hong Kong University of Science and Technology · 4 papers
- Massachusetts Institute of Technology · 4 papers
Use it
Download BibTeX of 22 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
- Efficient and adaptive thinking (231 papers)
- RL with verifiable rewards for reasoning (270 papers)
- Routing, pruning and distillation for reasoning (135 papers)
- Process reward models (39 papers)
Elsewhere in the topic
- Long CoT dynamics and efficiency (136 papers)
- Chain-of-thought training and analysis (127 papers)
- Efficient LLM serving (126 papers)
- Search and self-correction in reasoning (123 papers)
- Speculative decoding (92 papers)
- KV cache compression (45 papers)
- Theory of autoregressive reasoning (28 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).