deepseek-ai

/

DeepSeek-V3.2-Speciale

like 630

Follow

DeepSeek 113k

Added structured evaluation results from reasoning capabilities benchmarks:

• AIME 2025 (Pass@1): 96.0%
• HMMT 2025 (Pass@1): 99.2%
• HLE - Humanity's Last Exam (Pass@1): 30.6%
• Codeforces (Rating): 2701

These benchmarks evaluate the model's performance on:

• Mathematical reasoning (AIME 2025, HMMT 2025)
• General reasoning capabilities (HLE)
• Competitive programming (Codeforces)

This enables the model to appear in leaderboards and makes it easier to compare with other models.

Note: This PR adds benchmark metadata independently of PR #11 (which only updates model naming). Both changes affect different parts of the file and should be compatible.