Matharena
Evaluate LLMs on elite math competitions, from AIME to IMO.
MathArena is essential for researchers and model builders who need to stress-test mathematical reasoning at competition level. The leaderboard transparency and breadth of problem types make it superior to generic benchmarks, but the narrow focus and lack of API limit its use for non-specialists.
- AI researchers studying mathematical reasoning
- LLM developers fine-tuning for STEM tasks
- Competition math enthusiasts comparing model capabilities
- ML engineers evaluating open-source model math performance
- General-purpose LLM testing outside mathematics
- Non-technical users seeking simple chat-bot evaluations
- Users needing API access for automated benchmarking
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
In short
Matharena — Evaluate LLMs on elite math competitions, from AIME to IMO. Best for AI researchers studying mathematical reasoning, LLM developers fine-tuning for STEM tasks, Competition math enthusiasts comparing model capabilities. Free to use.
Viability Score
How likely is Matharena to still be operational in 12 months? Based on 4 signals — momentum (how recently it shipped), wrapper dependency, revenue model, and web presence.
Last calculated: July 2026
How we score →Key Features
- Interactive leaderboards by competition and date
- Final-answer problem benchmarks (AIME, HMMT, BRUMO, SMT, CMIMC, Apex)
- Proof-based competition benchmarks (USAMO, IMO, Putnam, Miklós Schweitzer)
- Visual math evaluation (Kangaroo problems grades 1-12)
- Project Euler problem set integration
- ArXivMath training dataset (June 2026)
- BrokenArXiv training dataset (June 2026)
- Per-cell raw output viewing for transparency
- Date-filtered performance snapshots
- Multiple competition seasons (2025, 2026)
- GLM 5.2 model on leaderboard (June 2026)
- Open leaderboard with community submissions
- Comparison tool for side-by-side model evaluation
About Matharena
MathArena is a benchmarking platform built to assess large language models on rigorous mathematical problems drawn from elite competitions. Developed by the SRI Lab at ETH Zurich and INSAIT, it tests genuine reasoning beyond standard multiple-choice or answer-extraction tasks. The platform covers final-answer competitions (AIME, HMMT, Apex Shortlist), proof-based competitions (USAMO, IMO, Putnam, Miklós Schweitzer), visual math (Kangaroo series), and Project Euler. In June 2026, two new training datasets—ArXivMath and BrokenArXiv—were released, and the leaderboard now includes GLM 5.2. Users browse interactive leaderboards segmented by competition and date, with raw model outputs viewable per cell. MathArena updates frequently as new competitions conclude, keeping benchmarks current. Its depth across competition types and transparency into outputs make it a reality check for AI mathematical reasoning, distinguishing it from broader benchmarks like MMLU or GSM8K.
Behind the Verdict
MathArena fills a gap left by standard benchmarks: it focuses on hard, multi-step reasoning rather than pattern matching. If you're fine-tuning a model for STEM, this is the closest thing to a real-world stress test. The inclusion of proof-based and visual math problems is rare among leaderboards, and the raw output inspection per cell is invaluable for diagnosing failures. Where it falls short is scope—it's strictly math competition problems, so don't expect evaluation on coding or general knowledge. For non-researchers, the UI is functional but utilitarian, and there's no API for automated benchmarking. Compared to Google's MMLU or OpenAI's evals, MathArena demands deeper reasoning; it's not a replacement but a complement. One caveat: the leaderboard is community-driven, so submission criteria are less controlled than a platform like Papers with Code. Still, for its niche, MathArena is the best public resource available.
Researching Matharena? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Use Cases
- Compare top LLMs on AIME 2026 to identify weaknesses in multi-step reasoning.
- Use BrokenArXiv dataset to fine-tune a model for robustness to erroneous problem statements.
- Evaluate a new open-weight model on proof-based IMO 2025 problems before publication.
- Analyze visual math performance on Kangaroo 2025 1-2 to assess multimodal reasoning.
- Track chronological performance trends in HMMT contests across 2025 and 2026.
- Benchmark a custom solver against Project Euler problems and share results.
Models Under the Hood
as of 2026-07-17
Limitations
- The platform is exclusively focused on mathematical competitions, so models are not tested on other domains.
- There is no mention of rate limits or plan gating, but as a free resource, reliance on server resources may cause sporadic slowdowns during peak usage.
Tools that pair well with Matharena
Common stack mates teams adopt alongside Matharena, with the specific reason each pairing earns its keep.
Featured Head-to-Head Comparisons
Alternatives to Matharena
View allFrequently Asked Questions
Categories
Best-of guides
Topics
Used Matharena? Help shape our editorial sentiment research.