Reasoning Benchmarks — Top AI Models

Benchmarks in this category

• MMLU — Massive Multitask Language Understanding — 57 academic and professional subjects.
• GPQA Diamond — Graduate-level Google-Proof Q&A in physics, chemistry, and biology.
• ARC-AGI — Abstraction and Reasoning Corpus — visual grid puzzles testing fluid intelligence.
• GSM8K — Grade-school math word problems requiring multi-step arithmetic reasoning.
• MGSM — Multilingual Grade School Math evaluates grade-school reasoning across ten languages.
• MMLU-Pro — Harder MMLU successor with 10-option questions and reduced contamination.
• BIG-Bench Hard — 23 challenging tasks from BIG-Bench where prior LMs underperformed humans.
• DROP — Discrete reasoning over paragraphs — math/counting/sorting inside reading comp.
• AGIEval — Human-centric standardized exams (SAT, GRE, GMAT, LSAT, civil service).
• MATH-500 — 500-problem competition-math subset used for reasoning model evals.
• AIME 2024 — American Invitational Math Examination — 15-problem olympiad set.
• AIME 2025 — The two 2025 AIME competition-mathematics forms used for exact-answer reasoning evaluation.
• FrontierMath — Expert-crafted research-level mathematics benchmark by Epoch AI.
• Humanity's Last Exam — Multi-disciplinary expert-level exam, contamination resistant.
• ARC-AGI 2 — Second-generation ARC Prize evaluation suite (2025).
• MuSR — Multistep soft reasoning over narrative scenarios.
• IFEval — Instruction-following evaluation with verifiable constraints.
• SimpleQA — Factuality benchmark of short single-fact questions.
• TruthfulQA MC2 — Truthfulness benchmark across 38 topic categories prone to misconceptions.
• Multilingual MMLU — MMLU translated across 26 languages by Cohere & contributors.
• Global-MMLU Lite — Culturally-balanced multilingual MMLU subset across 42 languages.
• FLORES-200 — Machine translation eval across 200 languages.
• WMT24 — WMT 2024 conference translation shared task.
• RULER 128K — Long-context evaluation across 13 synthetic tasks at 128K tokens.
• LongBench v2 — Realistic long-context tasks up to 2M tokens.
• ∞Bench — Tasks exceeding 100K tokens spanning code, math, and retrieval.
• LongFact Concepts — Long-form factuality benchmark measuring unsupported claims in open-ended concept answers.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Reasoning Benchmarks — Top AI Models.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path