Mapping the Open Model Landscape in 2025
summary
2025 is an inflection year: DeepSeek R1 drops in January and open Chinese labs (Qwen, Z.ai, Moonshot AI, Alibaba) release frequent models at scale (DeepSeek at ~600B–1T params), driving a widening gap with Western open models. Metrics show Quen passes Llama in cumulative adoption on Hugging Face while downloads prove noisy; Neatron nano 9B rivals Qwen at small sizes. The best fully reproducible release remains mo_232b (AI2) and the proposed Atom project budgets ~$100M to close capability and research gaps.
AI-written summary · about
chapters (24)
- 0:00Open model landscape intro
- 1:24Open model evolution
- 3:32Early 2025 model releases
- 4:13Chinese open models surge
- 6:21Download metrics limits
- 7:46Llama’s open source stance
- 9:11Companies adopt Chinese models
- 10:57Model adoption trends
- 12:40Benchmarks and Chinese dominance
- 13:45Chinese model providers
- 15:31Fine-tune trends on Hugging Face
- 16:54China’s open source strategy
- 18:59Local models and social concerns
- 20:04US tech leadership opportunities
- 21:07GPTOSS and open models
- 22:10Investment in fully open models
- 23:56Atom Project milestones
- 24:25Q&A: Open source vs weights
- 26:10Q&A: Open model influence
- 27:56Q&A: Sovereign AI in Europe
- 30:06Q&A: Model limitations
- 31:57Q&A: Fully open research models
- 32:19Q&A: Beyond language models
- 33:24Q&A: Gemma 3 challenges
related talks
54mThe Future of AI: Building the Most Impactful Technology Together
13mKeynote: To Nemotron and Beyond: Building the Open Source AI Ecosystem
44mApropos DeepSeek
36mInsights from The State of AI Report 2023
10mThe Future for AI-First Enterprises is “Open” - How to Succeed and Not...
25mGain full control of your apps with Open-Source AI
21mDemocratizing AI with Open Source Multi-Agent Systems
32mThe Rise of AI agents and open sources Ai models