Product Launchllmreasoningopen sourcemathematics
DeepSeek Releases V3.2 Open-Weight Models Suite
10.0
DeepSeek released two open-weight MIT-licensed models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, today; both are 685B-parameter, 690GB checkpoints and accompanied by a technical report. V3.2 runs on chat.deepseek.com, while Speciale was trained with extended reasoning-focused RL and DeepSeekMath-V2 rewards, achieving gold on the 2025 International Mathematical Olympiad benchmark. The release targets advanced reasoning, agent, and math workloads.
Key Points
- 1Announces two 685B-parameter MIT-licensed models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, 690GB each.
- 2Highlights Speciale's reasoning-focused RL training and inclusion of DeepSeekMath-V2 rewards to boost mathematical proofs.
- 3Enables practitioners to access open-weight models for advanced reasoning, agents, and math-intensive applications.
Scoring Rationale
Open-weight 685B release with rigorous RL and math training drives high impact, though wider benchmark validation is limited.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
High-Value Overnight OrdersEasyDelivered International ShipmentsMediumOn-Time Delivery Rate by CarrierHard
250 free problems · No credit card
See all Logistics & Shipping problems
