AlphaCode 2
Gemini-based competitive-programming system performing better than 85% of Codeforces participants, up from nearly 50% for AlphaCode.
Generates a large pool of candidate programs from Gemini-tuned models, then filters and ranks them. Shows heavy sampling plus verification lifts competitive programming from the median competitor to the top 15%.
- Date
- Wednesday, 6 December 2023
- Lab
- Google DeepMind
- Kind
- model
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| Codeforces participants outperformed | >85% AlphaCode (2022) at nearly 50% | company |
Tech report PDF could not be text-extracted; the 85% figure is from the Gemini launch post.
Sources
- blog.google/technology/ai/google-gemini-ai/
- storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf
This record was checked against its sources on 6 October 2026. How we check