According to emollick, OpenAI’s o1 graph shows no ability decay with more compute, hinting broad general LLM gains, per polynoamial’s launch comments.

According to @Smol_AI, DeepSeek has released V3.2 and 3.2-Speciale—state-of-the-art open weights language models positioned at GPT-5 capability levels, featuring significant advances in context management and transparency in compute scaling strategies (source: news.smol.ai, Dec 1, 2025). The models deliver detailed, well-documented research and new agent pipelines, including general, code, and search agents, enabling startups and enterprises to build sophisticated AI products without proprietary restrictions. These advancements present concrete business opportunities for AI-driven applications in software development, knowledge management, and automated search. DeepSeek’s open weights approach, combined with scalable infrastructure plans, increases accessibility for developers and reduces dependence on closed-source models, accelerating innovation in the AI industry.