AI News Feed
DeepSeek Open‑Sources 685B Models, Undercuts US APIs
03 Dec 2025- Chinese startup DeepSeek open-sources two 685B models (V3.2, V3.2‑Speciale) matching GPT‑5/Gemini, winning IOI medals, publishing weights on Hugging Face under MIT with much cheaper API pricing.
03 Dec 2025
Chinese AI startup DeepSeek has released V3.2 and V3.2‑Speciale, two 685B‑parameter reasoning models the company says match or exceed frontier systems like GPT‑5 and Gemini 3 Pro on math, tool use, and coding tests. DeepSeek claims the Speciale variant outperforms competitors in several areas and even posted gold‑medal scores at the 2025 International Math Olympiad and Informatics Olympiad (placing No. 10 overall at IOI).
Crucially, both models are distributed under an MIT license with weights available on Hugging Face, and DeepSeek is offering API pricing far below U.S. incumbents: V3.2 reportedly costs $0.28 input / $0.42 output per 1M tokens, vs. Gemini 3 Pro ($2 / $12), GPT‑5.1 ($1.25 / $10), and Sonnet 4.5 ($3 / $15). The release follows DeepSeek’s earlier R1 launch, which shook markets and triggered talks about U.S. chip export controls.
Why it matters: open‑sourcing near‑frontier models with competitive performance at a fraction of typical prices increases pressure on U.S. labs to justify premium API fees and could shift access and competition in the near‑frontier model space.
Source
Chinese AI startup DeepSeek has released V3.2 and V3.2‑Speciale, two 685B‑parameter reasoning models the company says match or exceed frontier systems like GPT‑5 and Gemini 3 Pro on math, tool use, and coding tests. DeepSeek claims the Speciale variant outperforms competitors in several areas and even posted gold‑medal scores at the 2025 International Math Olympiad and Informatics Olympiad (placing No. 10 overall at IOI).
Crucially, both models are distributed under an MIT license with weights available on Hugging Face, and DeepSeek is offering API pricing far below U.S. incumbents: V3.2 reportedly costs $0.28 input / $0.42 output per 1M tokens, vs. Gemini 3 Pro ($2 / $12), GPT‑5.1 ($1.25 / $10), and Sonnet 4.5 ($3 / $15). The release follows DeepSeek’s earlier R1 launch, which shook markets and triggered talks about U.S. chip export controls.
Why it matters: open‑sourcing near‑frontier models with competitive performance at a fraction of typical prices increases pressure on U.S. labs to justify premium API fees and could shift access and competition in the near‑frontier model space.
Source
The method
The prompts
Helping entrepreneurs and business owners use AI tools for growth and productivity.
Wawa Llama Ltd GB447340593