# πŸ“† ThursdAI - Qwen3-Coder & new A22B, πŸ‡ΊπŸ‡Έ AI action plan, LLMs win IMO, Sapient HRM & more AI news Source: https://www.youtube.com/watch?v=nLiabbM6cO4 Recap page: https://rapidrecap.app/video/nLiabbM6cO4 Generated: 2025-08-28T10:33:06.865+00:00 --- ## Quick Overview The "ThursdAI" episode for July 2024 highlights significant open-source AI model releases, particularly Alibaba's Qwen 3 and Qwen 3 Coder, which demonstrate state-of-the-art performance in various benchmarks, including coding and medical Q&A. The discussion also covers the US "Win AI Race" action plan, advancements in text-to-speech and diffusion models, and the surprising success of AI in math Olympiads, with OpenAI and DeepMind achieving gold medals. **Key Points:** - Alibaba released Qwen 3 (235B, 22B active parameters) and Qwen 3 Coder (480B, 35B active parameters), with the latter achieving state-of-the-art results on coding benchmarks like Sweetbench Verified, outperforming previous open-source models and rivaling proprietary ones. - The Qwen 3 (235B) model achieved the highest score on medical benchmarks (MedQA) among open-source models tested by the hosts, scoring 79.2 on Arena Hard and outperforming models like DeepSeek and Claude Opus in specific evaluations. - Alibaba's decision to release a non-reasoning Qwen 3 model, based on community feedback, yielded impressive results, demonstrating that models without explicit chain-of-thought reasoning can still achieve top-tier performance, even on tasks requiring reasoning. - The US White House released a "Win AI Race" action plan, focusing on deregulation and promoting AI development, which Joseph Nelson from RobFlow discussed in detail. - AI models from OpenAI and DeepMind achieved gold medals in the International Mathematical Olympiad (IMO), showcasing advanced reasoning capabilities that surprised mathematicians and highlighted the growing prowess of AI in complex problem-solving. - New advancements were noted in text-to-speech with Higs Audio V2 from Bzone AI, and in real-time diffusion with Mirage LSD from Deart AI, alongside research on subliminal learning in LLMs and Apple's multi-token prediction for faster inference. - The episode also covered the release of Mistral's Magistral model on Hugging Face and the Sapien Intelligence hierarchical reasoning model, a 27-million-parameter model showing significant results on tasks like Sudoku and mazes without pre-training. **Context:** This episode of "ThursdAI" features hosts Alex Vulov, Wolf from Raven Wolf, Yampel, Nistell Tahira, and LDJ discussing the latest developments in the AI landscape. The discussion centers on significant open-source model releases, particularly from Alibaba, and touches upon US AI policy, new research, and AI achievements in competitive fields like mathematics. ## Detailed Analysis The "ThursdAI" July 2024 episode dives deep into a week of major AI advancements, with a strong focus on open-source releases. Alibaba's Qwen 3 models, specifically the upgraded 235B version and the new 480B Qwen 3 Coder, are highlighted as benchmarks for open-source performance. The Qwen 3 (235B) model, noted for its non-reasoning architecture, achieved top scores on medical benchmarks and MMLU Pro, impressing the hosts with its capabilities despite the absence of explicit reasoning. The Qwen 3 Coder demonstrated state-of-the-art performance in coding tasks, rivaling proprietary models and achieving high scores on benchmarks like Sweetbench Verified. The hosts also discussed the US government's new "Win AI Race" action plan, which aims to foster AI development through deregulation, with insights provided by Joseph Nelson from RobFlow. Further advancements include new text-to-speech models like Higs Audio V2 and real-time diffusion models like Mirage LSD, alongside intriguing research papers on LLM behavior and inference speed. A notable event was the success of AI in the International Mathematical Olympiad, where OpenAI and DeepMind models secured gold medals, underscoring the rapid progress in AI's reasoning and problem-solving abilities. The discussion also touched upon Mistral's Magistral release and Sapien Intelligence's small, yet effective, hierarchical reasoning model. ### Open Source AI Releases - Alibaba's Qwen 3 (235B) and Qwen 3 Coder (480B) models showcased - Qwen 3 (235B) achieved top scores on medical benchmarks and MMLU Pro without explicit reasoning - Qwen 3 Coder set new state-of-the-art for open-source coding, rivaling proprietary models ### US AI Policy - The US White House released the "Win AI Race" action plan focused on deregulation and AI advancement - Joseph Nelson from RobFlow provided analysis on the policy's implications ### AI in Competitions & Research - OpenAI and DeepMind AI models won gold medals at the International Mathematical Olympiad - Research highlighted included Sapien Intelligence's small hierarchical reasoning model, Higs Audio V2 text-to-speech, Mirage LSD diffusion, subliminal learning, and Apple's inference speed improvements ### Model Architecture & Feedback - Alibaba's decision to release a non-reasoning Qwen 3 model was driven by community feedback for better control and performance - Discussion on the trade-offs between reasoning and non-reasoning models for various applications ### Sponsor Segment - Mention of Weights & Biases' day-one support for Qwen models and potential credits for users ### Other AI News - Release of Mistral's Magistral model - Discussion on the lack of effectiveness of Nvidia's Neotron reasoning models ### Key Quotes - "this new Quen model, I think it's around 480 billion. So, it's like a little bit less than half the size of of Kimmy's total size. Yet, it seems to be comparing comparatively, but if not like even better in a lot of areas." (LDJ) - "they removed reasoning from it and yet it performs incredibly." (Host) - "AI models from OpenAI and DeepMind achieved gold medals in the International Mathematical Olympiad" (Host)