MathNet-RAG - Embedding Retrieval: leaderboard

MathNet-RAG (2026) gives 35 Olympiad problems, each paired by experts with a related problem, and human experts grade the solutions. Accuracy when the model also gets one related problem and its solution found by an embedding retriever, in percent; higher is better.

Metric: on MathNet-RAG, 35 Olympiad problems each paired by experts with a related problem, graded by human experts; accuracy (%) with one related problem and its official solution retrieved by gemini-embedding-001; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview)92.9#123
2DeepSeek V3.2 Speciale89.5#277
3Grok 4.1 Fast83.8#431
4GPT-575.2#185
5Claude Opus 4.555.5#165
6OLMo 3 32B (Thinking)54.6
7Phi-4-reasoning-plus14.3#2288

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mathnet-rag-embedding-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-22.