AI Paper Atlas

Research map / Benchmarks / MMLU-STEM

MMLU-STEM leaderboard

10 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on MMLU-STEM. Best accuracy: 87 (GPT4-Code + CSV + voting, ICLR 2024).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

MMLU-STEM · accuracy

10 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
187GPT4-Code + CSV + voting
GPT4-Code
Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification
ICLR 2024 · MMLU results table
Agents, code and math
280.2w/ reranker + self-cons. + VR
Llama 3.1 8B Instruct
Reusing Pre-Training Data at Test Time is a Compute Multiplier
ICLR 2026 · Table 3
Reasoning and chain of thought
375Minerva 540B, maj1@k
PaLM 540B
Solving Quantitative Reasoning Problems with Language Models
NeurIPS 2022 · Table 2
Agents, code and math
472.3RSD (δ=0.7, PRM 7B)
Qwen2.5-Math-Instruct-7B target, 1.5B draft
Reward-Guided Speculative Decoding for Efficient LLM Reasoning
ICML 2025 · Table: Accuracy on reasoning b
Reasoning and chain of thought
566.8CompactDS (k=10)
Llama 3.1 8B Instruct
Frustratingly Simple Retrieval Improves Challenging, Reasoning-Intensive Benchmarks
ICLR 2026 · Table 1
Retrieval and knowledge
664.5MAmmoTH2-Plus-7B
Mistral-7B
MAmmoTH2: Scaling Instructions from the Web
NeurIPS 2024 · Main results table
Instruction and fine-tuning
758.3SLAT
DeepSeek-R1-Distill-Qwen-7B
SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoning
ICML 2026 · Table 5
Reasoning and chain of thought
849Llemma 34B
Code Llama 34B
Llemma: An Open Language Model for Mathematics
ICLR 2024 · Table 1
Agents, code and math
943.55MIND-OWM-14B Two Students
7B LLM
MIND: Math Informed syNthetic Dialogues for Pretraining LLMs
ICLR 2025 · Table: Results of 7B LLM train
Agents, code and math
1038.6SuCo
Qwen2.5-Math-1.5B
SuCo: Sufficiency-guided Continuous Adaptive Reasoning
ICML 2026 · Table 1
Reasoning and chain of thought

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.