Dr. Károly Zsolnai-Fehér reviews Qwen 3.8 Max and smaller variants, highlighting their 1.1M token context window, ability to run autonomous agents for up to 16 days, and scores above 50% on Humanity's Last Exam. Open models now cost far less than closed systems while matching performance.
This is insanity. We just got an amazing free AI system with Deepseek Flash on the lower end. Super quick, super cheap. But what about the open frontier models? This is Qwen 3.8 Max. And look, OpenAI and Anthropic are getting challenged there too. And not only that, but the price. Maybe even 5 to 10 times cheaper, depending. It is multimodal, so it has eyes and ears. 1 1 million token context window, and it is great for agentic workflows. This is actually one of the key points. They demonstrated it working independently while you get out there and live an active scholarly life. An absolutely lovely value proposition. You see, nobody has to be scared. It's not advertised by hacking into other people's systems, just productivity and tranquility. Man, sign me up for that. Especially that many other systems have to tap out within minutes to hours. And if you think 1 day of independent work is impressive, well, try 16! Yes, it sat there thinking for 16 days, starting from an empty folder, writing, testing, and repairing its own code. It can reproduce research papers and even improve them meaningfully. Ooh, stunning. It can also create incredible websites and apps with ease. The value proposition is so amazing, and the API pricing comparatively so low, I think it might force the other players to adjust their prices down. That is great for us Fellow Scholars. And they have committed to releasing the weights soon as well. Oho! Now, since it is gigantic, not many of us will be able to run the full model at home, but Here comes the best part. Dear Fellow Scholars, this is Two Minute Papers with Dr. Károly Zsolnai-Fehér. Now hold on to your papers, Fellow Scholars, because the best part is that there are going to be a variety of other smaller models too. The earlier smaller models, Qwen 3.6, 27, and 35 billion have legend status, and despite being several months old by now, many of you Fellow Scholars still consider them the best in their categories. They are the Toyota Corolla of the AI world. And that is the real news for most of us with more modest resources. Yup, this might be your next daily driver that you can own for free. Wow. Oh, and I almost forgot, I'll give you a little secret. There are many benchmarks, some of them are gamed, some of them less so. I think this is one of the good ones. Humanity's Last Exam. Devilishly difficult academic benchmark, when it started out, the best billion-dollar closed AI systems were able to do about 2% on it. Now, just a bit more than a year later, an open model—well over 50%. Woo-hoo! Keep an eye on this benchmark because it has been one of the most indicative of real-life performance for most brilliant Fellow Scholars like you. What a time to be alive! Kwan's contribution to open science and open source is simply incredible. We are being spoiled here with amazing new gifts every week. It is the golden age of open science and open AI systems. I am incredibly grateful for this. I use Lambda to reproduce AI research papers, often in minutes. It's also great to train your own models or fine-tune an existing one. Run inference, or text to image or video? Easy peasy. Running a DeepSeq chatbot or agent? Super fast, super reliable. Lambda gives you powerful NVIDIA GPUs to run your own experiments. I test ideas from the papers I cover, and moments later, results. Love it. Seriously, try it out now at lambda.ai/papers.
