Deepseek Reasoner: benchmark labels now distinguish AIME 2025 and Terminal-Bench 2.0

The OnAirToday AI Progress Tracker recorded a verified change to DeepSeek's Deepseek Reasoner (version DeepSeek-V3.2 (deepseek-reasoner, thinking mode)): Benchmark labels now distinguish AIME 2025 and Terminal-Bench 2.0.

DeepSeek Reasoner’s documented DeepSeek-V3.2 generation combines text-based reasoning with tool use for mathematics, coding and multistep workflows. Published evaluations show strong but imperfect performance, while the mutable API alias and hosted-service privacy risks warrant careful version tracking and deployment controls.

What changed in this review:

• Aider Polyglot and OTIS figures are excluded because attribution to this generation is not established.
• Added the January 2025 DeepSeek database exposure, explicitly distinguished from a V3.2 model vulnerability.
• Capability score 70 → 66
• Flaws and limitations updated
• Risks and threats updated

Anyone building on Deepseek Reasoner should re-check the model's risk dashboard before shipping, since flaws and reported incidents are what moved in this review.

The full profile — capabilities, benchmark citations, known flaws and risk guidance — is on the tracker at /ai-models/deepseek-reasoner.

References:

• https://github.com/deepseek-ai/DeepSeek-V3.2
• https://api-docs.deepseek.com/quick_start/pricing
• https://api-docs.deepseek.com/guides/reasoning_model
• https://www.wiz.io/blog/wiz-research-uncovers-exposed-deepseek-database-leak