Frontier AI Models: A Review of Past Nine Months
summary
Survey finds reasoning models like o3 and DeepSeek R1 move workflows from iterative chat to single‑shot problem solving by fine‑tuning on stepwise reasoning and using chain of thought prompting; open source variants match proprietary accuracy and enable deployment that fits on 1–2 H100s after quantization. Reported cost falls from ~$60 to ~$2 per million tokens — roughly 10–30x cheaper inference — and Llama 4 adds multimodal capability that lets enterprises replace proprietary OCR.
AI-written summary · about
chapters (15)
- 0:00Frontier AI models overview
- 3:53AI models state Oct 2024
- 5:57Reasoning models emerge Jan 2025
- 7:44Chain of thought prompting
- 9:29How reasoning models generalize
- 10:53OpenAI O1 reasoning leap
- 11:37DeepSeek R1 open source
- 12:21Llama 4 multimodal models
- 13:26AI progress Aug 2025
- 14:08Four AI model changes
- 15:50Q&A
- 17:54Reasoning vs tool calling
- 19:42Open source adoption concerns
- 22:12Future AI architectures
- 23:38Enterprise adoption challenges
related talks
24mThe Rise of Reasoning Models
19mRedefining Intelligence: How Reasoning Is Re-Shaping AI in 2025
18mTrends Across the AI Frontier
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence
54mThe Future of AI: Building the Most Impactful Technology Together
14mReliable Agentic AI, the Open Way
34mMapping the Open Model Landscape in 2025
31mScaling AI for Industry