Research map / Large language models / Agents, code and math
Reasoning benchmarks and platforms
Probe LLM reasoning with games, multilingual math, and competitive programming evaluations.
47 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Agents, code and math, keyword cluster “evaluating · capabilities · bench”
State History Tried 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.
6 papers in 2023–24 and 39 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.7% to 3.1%.
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
- 2016 · QART: A System for Real-Time Holistic Quality Assurance for Contact Center Dialogues (AAAI 2016 · 16 citations)
- 2018 · Learning Combinatory Categorial Grammars for Plan Recognition (AAAI 2018 · 9 citations)
- 2023 · Abstract Visual Reasoning: An Algebraic Approach for Solving Raven's Progressive Matrices (CVPR 2023 · 9 citations)
- 2024 · Position: LLMs Can’t Plan, But Can Help Planning in LLM-Modulo Frameworks (ICML 2024 · 11 citations)
- 2025 · Absolute Zero: Reinforced Self-play Reasoning with Zero Data (NeurIPS 2025 · 7 citations)
- 2026 · Evaluating Relational Reasoning in LLMs with REL (ICML 2026 · 0 citations)
Earliest papers here
- QART: A System for Real-Time Holistic Quality Assurance for Contact Center Dialogues (AAAI 2016 · 16 citations)
- Learning Combinatory Categorial Grammars for Plan Recognition (AAAI 2018 · 9 citations)
- Abstract Visual Reasoning: An Algebraic Approach for Solving Raven's Progressive Matrices (CVPR 2023 · 9 citations)
What has been tried
30 papers here report 268 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 7%, hurt 17%, 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 |
|---|---|---|---|---|
| Train on deduction tasks only, removing abduction and induction | OOD math benchmarks · Math average | 38.4 -> 32.0 | Hurt | Absolute Zero: Reinforced Self-play Reasoning with Zero Data (NeurIPS 2025 · 7 citations) |
| Add KPMath-G to MixMath | GSM8K · accuracy | 75.7 -> 80.7 | Helped | Key-Point-Driven Data Synthesis with Its Enhancement on Mathematical Reasoning (AAAI 2025 · 4 citations) |
| remove CoT filtering | MVBench · accuracy | 55.6 -> 53.7 | Hurt | Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation (CVPR 2025 · 4 citations) |
| Add iterative online DPO training | Complex Problem-Solving · Avg. | 42.1 -> 45.4 | Helped | WebThinker: Empowering Large Reasoning Models with Deep Research Capability (NeurIPS 2025 · 4 citations) |
| Use 2-shot Chain-of-Thought prompting instead of zero-shot IO prompting | ACPBench mean Bool · accuracy | 87.02 -> 89.59 | Mixed | ACPBench: Reasoning About Action, Change, and Planning (AAAI 2025 · 3 citations) |
| Add ReAct-specific prompting to reduce early stopping | Movie Recommendation · accuracy | 68.60 -> 72.47 | Helped | An LLM Compiler for Parallel Function Calling (ICML 2024 · 3 citations) |
| Add the full reward function, including format reward | BIRD Dev · accuracy | 60.4 -> 63.1 | Helped | SQL-R1: Training Natural Language to SQL Reasoning Model By Reinforcement Learning (NeurIPS 2025 · 1 citation) |
| Prompt tune Llama2 7b on RMath using the problem-solving framework | RMath · accuracy | 0 -> 18.5 | Helped | RMath: A Logic Reasoning-Focused Datasets Toward Mathematical Multistep Reasoning Tasks (AAAI 2025 · 0 citations) |
| replace weak-to-strong control with naive logit addition | MMToM-QA · accuracy (all) | 79.38 -> 74.45 | Hurt | Overcoming Multi-step Complexity in Multimodal Theory-of-Mind Reasoning: A Scalable Bayesian Planner (ICML 2025 · 0 citations) |
| add a calculator | MapEval-Textual straight-line distance questions · accuracy | 51.06% to 85.11% | Helped | MapEval: A Map-Based Evaluation of Geo-Spatial Reasoning in Foundation Models (ICML 2025 · 0 citations) |
| Use chain-of-thought prompting instead of non-CoT prompting | BMMR-Eval · overall performance | Helped | BMMR: A Large-Scale Bilingual Multimodal Multi-Discipline Reasoning Dataset (NeurIPS 2025 · 0 citations) | |
| increase candidate pool |D| from 4x to 8x batch size for LILO on VinePPO | GSM8K · test accuracy (%) | further improved over |D|=4x (exact value not given in text) | Helped | LILO: Learning to Reason at the Frontier of Learnability (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 28 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 · 2 of 28 papers
- MATH · 2 of 28 papers
- HLE · 2 of 28 papers
- AMC · 2 of 28 papers
- AIME · 2 of 28 papers
Baselines
- GPT-4o · 4 of 28 papers
- Gemini 2.5 Pro · 4 of 28 papers
- DeepSeek-R1-671B · 3 of 28 papers
- DeepSeek-R1 · 3 of 28 papers
- o4-mini · 3 of 28 papers
- GPT-4o Mini · 2 of 28 papers
- GPT-4 · 2 of 28 papers
- WizardMath · 2 of 28 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 45 papers; from OpenAlex and the papers' affiliations).
- University of Oxford · 5 papers
- Microsoft · 4 papers
- Shanghai Jiao Tong University · 3 papers
- Tsinghua University · 3 papers
- Fudan University · 3 papers
- IBM · 2 papers
- University of Science and Technology of China · 2 papers
- University of California, Berkeley · 2 papers
- Shanghai AI Laboratory · 2 papers
- Zhejiang University · 2 papers
Use it
Download BibTeX of 23 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
- Agent instruction-following benchmarks (176 papers)
- Long-horizon LLM agents (155 papers)
- Software engineering agents (140 papers)
- LLM-driven embodied and driving agents (54 papers)
Elsewhere in the topic
- Mathematical and structural reasoning benchmarks (304 papers)
- Inductive and abstract reasoning evaluation (158 papers)
- LLMs for optimization and heuristics (147 papers)
- LLMs for scientific discovery (100 papers)
- Code generation benchmarks (96 papers)
- LLM-guided program synthesis (93 papers)
- Execution-based code verification (90 papers)
- Retrieval-augmented formal theorem proving (62 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).