---
title: "Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad"
authors:
  - "Google DeepMind"
year: 2025
venue: "DeepMind blog (IMO-certified)"
arxiv: null
doi: null
source: "https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/"
topics:
  - "mathematical-reasoning"
  - "imo"
seed_rank: 1230
seed_batch: "alpoge-math-2026-08-24"
reviewed: "2026-08-24"
pool: "agents"
relevance_score: 10
lineage: ai-mathematical-reasoning
cites:
  []
see:
  - "1229-openai-experimental-reasoning-model-imo-2025-gold-cutoff-proof"
  - "1223-olympiad-level-formal-mathematical-reasoning-with-reinforcemen"
  - "1222-gold-medalist-performance-in-solving-olympiad-geometry-with-al"
---

# Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad

## One-sentence takeaway

Gemini Deep Think, end-to-end natural language, contest time: 35/42 on IMO 2025, official IMO graders, gold-medal standard — the certified twin of OpenAI's 35/42 announcement.

## Why it matters here

2025 gold, official. Unlike 2024 AlphaProof (Lean, days, expert formalization), this is informal proofs from the official statements inside 4.5 hours.

## Problem

Silver 2024 was formal and slow. Can a general model do IMO in natural language on the clock?

## Design

Deep Think: parallel exploration of solution candidates rather than a single chain. No tools. Same constraints as humans.

## Evidence

DeepMind blog, July 2025. IMO graders: "clear, precise and most of them easy to follow." Solutions released. Same 35/42 as OpenAI's experimental model (1229).

## Limitations

"Advanced version" not the public Gemini 2.5 Pro (which scored far lower without the pipeline). P6 unsolved. Blog, not a Nature paper (yet).

## Implications for Broadside

The certified 2025 gold. See 1223 for 2024 silver and 1229 for OpenAI's parallel claim.

## Bottom line

Mint the DeepMind blog. Do not remint as AlphaProof.

## Links

- https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/