Leverage DeepSeek-V3.2-Speciale for Competition-Level Math Reasoning
Access the Speciale endpoint for world-class mathematical reasoning performance.
Use the temporary 'Speciale' endpoint (base_url="https://api.deepseek.com/v3.2_speciale_expires_on_20251215") for elite-level math tasks that general models fail to solve.
The Scenario
You are preparing educational materials or solving elite-level competitive mathematics problems that require state-of-the-art reasoning capabilities.
Before & after
Users relied on general-purpose models for math, often resulting in errors on complex Olympiad-level problems. Cost: Hours of manual verification.
Switching to the DeepSeek-V3.2-Speciale endpoint allows you to access IMO-gold-medal level math reasoning. Cost: 2 minutes to switch the base_url in your script.
The Prompt
Change your API base URL to: https://api.deepseek.com/v3.2_speciale_expires_on_20251215 Then send your math query: [PASTE_COMPLEX_MATH_PROBLEM_HERE]
Until December 15, 2025, DeepSeek is hosting a specialized 'Speciale' model. This model is exceptionally strong at mathematical reasoning, achieving near gold-medal International Mathematical Olympiad (IMO) performance. Note that it does not support tool calls.
Source
Change Log | DeepSeek API Docs"Speciale... achieved an impressive 35 out of 42 points on the International Mathematical Olympiad benchmark - approaching IMO gold medal performance."