RedNote, the Shanghai-based company behind the social platform Xiaohongshu, has made history by developing the world's first artificial intelligence model to achieve a perfect score at the International Mathematical Olympiad (IMO). The model, dots-note-3.0, solved all six problems and secured a flawless 42 out of 42 points at this year's competition, which concluded in Shanghai on Monday, 21 July 2026.

A historic milestone in AI reasoning

The IMO, held annually since 1959, is widely regarded as the world's most prestigious mathematics competition — and increasingly a benchmark for advanced AI reasoning. Unlike multiple-choice or answer-only formats, IMO problems demand rigorous proof-based solutions, free of logical errors or missing conditions, making a perfect score an exceptionally demanding target for any system.

Notably, dots-note-3.0 is described as the lightest model in RedNote's dots3 family, which also includes two larger variants — jazz and aria — designed for different use cases and compute costs. The achievement is therefore especially striking given the model's comparatively compact architecture.

Why it matters

The result marks a decisive leap beyond what the industry's largest players accomplished just twelve months ago. At IMO 2025, both Google DeepMind and OpenAI reached gold-medal-level performance, each scoring 35 out of 42 — strong results, but still 7 points short of perfection. RedNote's dots-note-3.0 has now cleared that gap entirely, resetting expectations for what frontier reasoning models can achieve.

Mathematical olympiad performance is considered a proxy for deep, multi-step logical reasoning — a capability that underpins applications ranging from scientific research automation to code verification and formal theorem proving.

Open-source plans and competitive backdrop

According to a post on the model's official website, RedNote said it would open-source dots-note-3.0 'in the near future', without providing a specific timeline. If released, the move would give researchers and developers worldwide direct access to the model's weights, potentially accelerating downstream applications in mathematics, science, and engineering.

The announcement arrives amid an intensifying global race among AI labs — particularly in China — to demonstrate frontier reasoning capabilities. Companies including DeepSeek, Alibaba, and Baidu have all released competitive open-weight models in recent months, and RedNote's entry signals that the competitive field is broadening well beyond dedicated AI research organisations.

What's next

The dots-note-3.0 model remains in beta phase as of publication. The open-source release, once it materialises, will be a key moment to watch: it will determine whether the model's IMO-level reasoning translates into practical gains for the wider developer community. Analysts and researchers will also be scrutinising whether Google DeepMind, OpenAI, and other incumbents respond with updated reasoning benchmarks or model releases of their own.