# 📅 ThursdAI - GPT5 is here Source: https://www.youtube.com/watch?v=e6kwtQadPk0 Recap page: https://rapidrecap.app/video/e6kwtQadPk0 Generated: 2025-08-28T10:30:34.458+00:00 --- ## Quick Overview OpenAI released open-source models GPT OSS 120B and 20B, marking a significant shift towards open AI development, alongside releases from other labs like Quen and Hanyan, while Anthropic updated its Opus model to 4.1, and discussions focused on the practical applications and performance of these models, especially concerning local execution on devices. **Key Points:** - OpenAI released GPT OSS 120B and 20B models under an Apache 2.0 license, making them openly available and highly performant, with the 20B model being particularly praised for its accessibility and speed. - The Quen 3 models, specifically the 4B instruct and thinking models with 256k context, were highlighted as revolutionary for mobile devices, running efficiently on iPhones and offering significant capabilities for edge computing. - Anthropic updated its top-tier model to Claude Opus 4.1, noting improvements in coding and reasoning, though it remains a premium option, with users preferring it for complex coding tasks. - The community discussed the trade-offs between models trained on vast internal knowledge versus those focusing on tool-use and external data retrieval, with a leaning towards tool-calling for up-to-date information and reduced copyright concerns. - Discussions around hardware for running local models revealed a diverse range, from high-end AI workstations with multiple GPUs to consumer-grade MacBooks, emphasizing the increasing feasibility of local AI execution. - The release of GPT OSS is seen as a major gift to the development world, with its Apache 2.0 license and detailed documentation, fostering community engagement and further innovation in open-source AI. - Concerns were raised about the models' potential over-censorship and limitations in multilingual capabilities and creative writing, with some users expressing disappointment that GPT OSS wasn't a broader, more general-purpose model like GPT-4. **Context:** This episode of ThursdAI covers a significant week in AI, dominated by the highly anticipated release of OpenAI's open-source models, GPT OSS 120B and 20B. The discussion features insights from AI enthusiasts and developers, including Ryan Carson, exploring the technical details, performance benchmarks, and practical implications of these new models, as well as concurrent releases from other AI labs and updates to existing powerful models like Anthropic's Claude Opus. ## Detailed Analysis The week's AI news centers on OpenAI's release of GPT OSS 120B and 20B under an Apache 2.0 license, a move celebrated for making advanced AI more accessible. The 20B model, in particular, is lauded for its performance and ability to run locally on consumer hardware, including mobile devices like iPhones, with its 256k context length being a standout feature. Quen's 3 4B models also garnered attention for their mobile-friendly design and efficiency. Anthropic's Claude Opus 4.1 received updates, maintaining its position as a preferred model for complex coding tasks, though its cost remains a barrier for daily use. The conversation delved into the future of AI models, debating the merits of embedding extensive knowledge versus relying on tool-use for real-time data, with a growing preference for tool-calling due to its benefits in data currency and avoiding copyright issues. Hardware discussions highlighted the increasing viability of local AI processing, with participants sharing their setups from high-end workstations to everyday laptops. Despite the excitement, critiques emerged regarding potential over-censorship, limitations in multilingual support, and a perceived lack of broad general-purpose capabilities in the OpenAI releases, leading to mixed user reactions. ### OpenAI's Open Source Release - GPT OSS 120B and 20B models launched under Apache 2.0 license - 20B model praised for local performance and accessibility on devices - OpenAI's shift from closed to open AI celebrated by the community ### Other Open Source Models - Quen 3 4B instruct and thinking models highlighted for mobile efficiency and 256k context - Hanyan models released in various small sizes - Xay04 from China outperforming OpenAI's 03 mini in benchmarks ### Anthropic Updates - Claude Opus 4.1 released with improvements in coding and reasoning - Opus remains a preferred choice for complex coding tasks despite cost - Benchmark scores for Opus 4.1 show incremental gains ### Model Development Philosophies - Debate on embedding knowledge vs. tool-use - Preference for tool-calling for data currency and reduced copyright concerns - Discussion on model quantization and its impact on performance ### Hardware for AI - Diverse user hardware discussed, from high-end AI workstations to MacBooks - Emphasis on increasing feasibility of running AI models locally - Importance of GPU availability and VRAM for model performance ### Community Reception and Criticisms - Mixed reactions to GPT OSS, with some finding it over-censored or not a generalist model - Concerns about multilingual capabilities and creative writing performance - Discussion on the value of models acting as archives of knowledge ### Future AI Expectations - Desire for models excelling in coding and multilingual support - Speculation on GPT-5's capabilities, particularly in coding and AGI advancements - Importance of community feedback in model development