# AI News: 5 New Models Dropped This Week! Source: https://www.youtube.com/watch?v=5cMZqjrgq6Y Recap page: https://rapidrecap.app/video/5cMZqjrgq6Y Generated: 2026-02-20T17:44:42.69+00:00 --- ## Quick Overview The past week saw major AI announcements including Anthropic releasing Claude Sonnet 4.6 with a 1M token context window, Google launching Gemini 3.1 Pro which shows significant gains in reasoning and coding benchmarks over Gemini 3 Pro, a new open-weight model Qwen 3.5-397B-A17B from Alibaba, ByteDance's Seedance 2.0 video model facing backlash from Hollywood for unauthorized use of IP, and Meta patenting an AI system to simulate deceased users' social media activity. **Key Points:** - Anthropic released Claude Sonnet 4.6, featuring improved skills in coding and design, and a 1M token context window in beta (0:14). - Google launched Gemini 3.1 Pro, which outperforms Gemini 3 Pro and approaches GPT-4/Opus levels in benchmarks like ARC-AGI-2 (77.1% vs 58.3%) and complex reasoning tasks (8:05). - Alibaba released Qwen 3.5-397B-A17B, the first open-weight model in the Qwen 3.5 series, which is natively multimodal and Apache 2.0 licensed (19:23). - ByteDance's Seedance 2.0 AI video model faced backlash from Hollywood studios (MPA, SAG-AFTRA) for allegedly using copyrighted material without safeguards (21:13, 21:36). - Meta patented an AI system capable of simulating a deceased user's social media activity, including continuing to post (28:13). - Bytedance announced it will add safeguards to Seedance 2.0 following the Hollywood backlash (22:47). - New models from other labs include Seed2.0 from Bytedance and the open-source Phoenix-4 real-time human rendering model from Tavus (18:12, 26:53). ![Screenshot at 0:09: The video begins by highlighting the introduction of Claude Sonnet 4.6, Anthropic's new model, shown on their blog post titled 'Introducing Claude Sonnet 4.6'.](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-00-09.jpg) **Context:** This video covers major Artificial Intelligence news from the past week, focusing on new model releases from key players like Anthropic and Google, as well as controversies involving intellectual property and the ethical implications of AI, such as Meta's patent on simulating deceased users and Hollywood's reaction to Bytdance's video generation model, Seedance 2.0. ## Detailed Analysis The AI news recap for the week covers several significant model releases and industry developments. Anthropic released Claude Sonnet 4.6, noting it is their most capable Sonnet model yet, featuring upgrades across coding, reasoning, and design, and notably including a 1M token context window in beta (0:14, 0:45). For users on free/pro plans, Sonnet 4.6 is now the default model at claude.ai and Claude Cowork, maintaining the same pricing as Sonnet 4.5 (10:10). Google launched Gemini 3.1 Pro, described as a smarter model for complex tasks, showing significant performance gains over Gemini 3 Pro, especially in reasoning benchmarks like ARC-AGI-2 (77.1% vs 51.1%) and coding benchmarks like GPQA Diamond (96.2% vs 91.9%) (7:26, 7:55). The speaker also mentioned that Gemini 3.1 Pro is now available in Google AI Studio (9:33). From open-weight models, Alibaba released Qwen 3.5-397B-A17B, an open-weight, multimodal model licensed under Apache 2.0, featuring 8.6x-19.0x decoding throughput improvements over Qwen 3-Max (19:15). In terms of controversy, ByteDance's Seedance 2.0 video generation model faced backlash from Hollywood groups like the MPA and SAG-AFTRA for unauthorized use of copyrighted works and likenesses, prompting ByteDance to promise safeguards (21:13, 22:51). Finally, Meta patented an AI system capable of simulating a deceased person's online behavior, including continuing to post, which the speaker found dystopian and creepy (28:05). The speaker also highlighted Tavus's Phoenix-4, a real-time human rendering model that shows improved emotional responsiveness compared to previous iterations (26:53). ### Anthropic Claude Sonnet 4.6 - Most capable Sonnet model yet - Full upgrade across coding, computer use, reasoning, agent planning, knowledge work, and design - Features a 1M token context window in beta (0:14, 0:45) - Pricing remains the same as Sonnet 4.5 ($3/$15 per million tokens) (10:10) ### Google Gemini 3.1 Pro - Smarter model for complex tasks - Outperforms Gemini 3 Pro on benchmarks like ARC-AGI-2 (77.1%) and GPQA Diamond (96.2%) (7:55) - Available via Gemini API in Google AI Studio, Gemini CLI, Vertex AI, and Gemini Enterprise for developers/enterprises (7:32) ### Alibaba Qwen 3.5-397B-A17B - First open-weight model in Qwen 3.5 series - Native multimodal, trained for real-world agents - Powered by hybrid linear attention + sparse MoE - Apache 2.0 licensed (19:15) ### Industry Controversy (Seedance 2.0) - ByteDance's Seedance 2.0 faced backlash from MPA/SAG-AFTRA for unauthorized use of copyrighted works, leading ByteDance to promise safeguards (21:13, 22:51) ### Meta's Digital Afterlife Patent - Meta patented an AI system to simulate a deceased user's online activity, including posting, which the speaker found dystopian (28:05) ### Other AI Updates - ByteDance's Seedance 2.0 video model released (21:13) - Tavus Phoenix-4 real-time human rendering model shows improved emotional realism (26:53) - ByteDance's new model is open-weight (19:15) ![Screenshot at 0:00: Host speaking directly to the camera in a home studio setup, holding a mug.](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-00-00.jpg) ![Screenshot at 0:10: Anthropic blog post announcing 'Introducing Claude Sonnet 4.6' with an abstract network graphic.](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-00-10.jpg) ![Screenshot at 1:14: Chart showing the increasing performance of Claude Sonnet models over time on 'Computer use' benchmarks, culminating in Sonnet 4.6 at 72.5%.](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-01-14.jpg) ![Screenshot at 1:27: A table comparing various Claude models \(Sonnet 4.6, Opus 4.5, GPT-4, etc.\) across numerous benchmarks, highlighting performance metrics.](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-01-27.jpg) ![Screenshot at 2:10: A detailed comparison chart showing Seed 2.0 scores against competitors like Opus 4.5 and GPT-5.2 on various benchmarks like Visual Reasoning & Perception \(77.1% for Seed 2.0 on ARC-AGI-2\) \(2:04\).](https://ss.rapidrecap.app/screens/5cMZqjrgq6Y/00-02-10.jpg)