Summary
In this Two Minute Papers episode, Dr. Károly Zsolnai-Fehér covers the arrival of Qwen 3.8 Max, a new large frontier model that is challenging OpenAI and Anthropic on both capability and cost. The model is fully multimodal — supporting vision and audio — offers a 1-million-token context window, and is priced at an estimated five to ten times cheaper than comparable closed models, a gap Dr. Zsolnai-Fehér believes may force competitors to lower their own pricing.
A standout demonstration involved the model running autonomously for 16 consecutive days from an empty folder: writing, testing, and repairing its own code, reproducing existing research papers, and improving on them. The video also highlights the broader Qwen open-weight family — the 3.6B, 27B, and 35B parameter variants — which Dr. Zsolnai-Fehér describes as the “Toyota Corolla of AI” for their durable, class-leading performance despite being several months old.
For benchmark context, the episode points to Humanity’s Last Exam as one of the more reliable and harder-to-game evaluations. When the benchmark launched, the best closed commercial models scored around 2%; an open model has now crossed 50% — a striking illustration of how quickly open-source AI is closing the gap on frontier systems. Lambda is featured as a sponsor providing GPU compute for reproducing AI research. The episode is a concise entry point for anyone tracking the competitive dynamics between open-weight and closed AI model providers.
📺 Source: Two Minute Papers · Published August 05, 2026
🏷️ Format: News Analysis
