DeepSeek open-sources math reasoning model that reaches IMO gold-level performance

Summary

China’s DeepSeek released DeepSeekMath‑V2, an open-weights model focused on self-verifiable mathematical reasoning and theorem proving. The model reports gold‑medal–level results on the 2025 International Mathematical Olympiad, adding open access to a capability previously shown by proprietary systems; this could lower barriers for researchers and developers building rigorous reasoning agents.