OpenAI reports 12 of 12 problems and Google DeepMind 10 at ICPC World Finals 2025

OpenAI and Google DeepMind announced their results from the 2025 ICPC World Finals in Baku on 2025-09-17 (the Gigazine summary below is dated 2025-09-18; the finals were held earlier in September and…

Date
17 September 2025
Who
OpenAI, Google DeepMind
Confidence
Medium (company-reported; conditions differed)
Deep dive
Reasoning II, from o1 and o3 to DeepSeek-R1 and the labs that replicated them

Tier: Supporting · Significance: 3/5 · Org(s): OpenAI, Google DeepMind · Confidence: Medium (company-reported; conditions differed) OpenAI and Google DeepMind announced their results from the 2025 ICPC World Finals in Baku on 2025-09-17 (the Gigazine summary below is dated 2025-09-18; the finals were held earlier in September and I did not verify the contest day). OpenAI reported solving all 12 problems, with GPT-5 solving 11 on the first attempt and an experimental reasoning model solving the last after nine submissions, on-site under the five-hour limit. Google DeepMind's enhanced Gemini 2.5 Deep Think, participating online, solved 10, including one that no human team solved, with eight inside the first 45 minutes. The best human team solved 11 and the University of Tokyo, second among humans, 10 (Gigazine summary). The open-weight DeepSeek-V3.2-Speciale later reported 10 of 12 on the same contest, plus 492/600 at IOI 2025 (V3.2 paper, Table 4). One caveat is that no wrong-submission penalty was applied to OpenAI's runs in the on-site AI environment (reported), and compute budgets are undisclosed. It is a companion to B08-34; see also B10.

Read it in the deep dive