# This Week in All Things AI - Week 49-2025

> Sunday 30th November 2025 to Saturday 6th December 2025

**Published by:** [This Week in All Things AI](https://paragraph.com/@twiata/)
**Published on:** 2025-12-07
**URL:** https://paragraph.com/@twiata/this-week-in-all-things-ai-week-49-2025

## Content

Week 49 was relatively quiet with just Deepseek-3.2 being the big model announcement. Rumors are strong that Week 50 might have GPT-5.2 from OpenAI and Grok-4.20 from xAI released Below is my personal website which aggregates links to many of my socials as well as content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to themgoolamabbas | Instagram, TikTok | LinktreeChief Knowledge Officer @ Animoca Brands. Also assists in Growth Marketinghttps://linktr.eeThe following messages were posted on the 'All Things AI' Telegram group from Sunday 30th Nov 2025 to Saturday 6th Dec 2025Sunday 30th November 2025Monday 1st December 2025Circled around the sun one more time. Thank you to everyone for the wishesTuesday 2nd December 2025DeepSeek Unveils Powerful V3.2 AI Models with Top Competition Scores Deepseek released DeepSeek-V3.2 and V3.2-Speciale on Monday, building on prior work with efficient attention tech and heavy reinforcement learning. They claim gold-medal wins in 2025 math and coding olympiads like IMO and IOI, plus top scores on AIME and GPQA. V3.2 is fully open-source on Hugging Face and available via app or API at low costs—about 4-30 times cheaper than rivals like GPT-5. V3.2-Speciale, a reasoning-focused version, is API-only for now, with real-world tests still ahead. DeepSeek @deepseek_ai Launching DeepSeek-V3.2 & DeepSeek-V3.2-Speciale — Reasoning-first models built for agents! DeepSeek-V3.2: Official successor to V3.2-Exp. Now live on App, Web & API. DeepSeek-V3.2-Speciale: Pushing the boundaries of reasoning capabilities. API-only for now. Tech 15.4K 7:19 PM • Dec 1, 2025 Chubby♨️ @kimmonismus DeepSeek V3.2(special) is a massive release! I think some people don't understand just how massive this release is! - They are the first, even ahead of OpenAI and Google, to release a Gold IMO 2025, CMO 2025, IOI 2025, and ICPC World Finals model! Everyone now has access to Chubby♨️ @kimmonismus The whale is back! Holy moly, look at those evals, sota, even outperforming Gemini 3.0 pro and GPT-5 high on several benchmarks. They were cooking! V3.2: Balanced inference vs. length. Your daily driver at GPT-5 level performance. V3.2-Speciale: Maxed-out reasoning 1,912 8:47 PM • Dec 1, 2025 Cline @cline Live in Cline, DeepSeek-V3.2 & V3.2-Speciale. V3.2 is near GPT-5 level, while Speciale rivals Gemini-3.0-Pro. $0.28/$0.42 per million tokens, 131K context window. DeepSeek @deepseek_ai Launching DeepSeek-V3.2 & DeepSeek-V3.2-Speciale — Reasoning-first models built for agents! DeepSeek-V3.2: Official successor to V3.2-Exp. Now live on App, Web & API. DeepSeek-V3.2-Speciale: Pushing the boundaries of reasoning capabilities. API-only for now. Tech 249 1:37 AM • Dec 2, 2025 Wednesday 3rd December 20251Anthropic acquires Bun the speedy JavaScript runtime, bundler, and package manager popular with Netflix and Spotify. This follows Claude Code, their AI coding assistant launched in May 2025, reaching $1 billion in annualized revenue just six months later. The Bun team joins Anthropic to improve Claude Code, the Agent SDK, and future products, while keeping Bun open source. Chief Product Officer Mike Krieger praised Bun's technical excellence after months of partnership. Mike Krieger , many might remember co-founded Instagram with Kevin Systrom and was Instragram's CTO prior to its acquisition by Meta Bun @bunjavascript Bun is joining Anthropic! bun.com Bun is joining Anthropic Bun has been acquired by Anthropic. Anthropic is betting on Bun as the infrastructure powering Claude Code, Claude Agent SDK, and future AI coding products & tools. 11.7K 2:03 AM • Dec 3, 2025 Anthropic @AnthropicAI Anthropic is acquiring @bunjavascript to further accelerate Claude Code’s growth. We're delighted that Bun—which has dramatically improved the JavaScript and TypeScript developer experience—is joining us to make Claude Code even better. Read more: anthropic.com Anthropic acquires Bun as Claude Code reaches $1B milestone Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems. 9,544 2:01 AM • Dec 3, 2025 2Mistral launches Mistral 3, a family of 10 models under the Apache 2.0 license, including its new flagship Mistral Large 3 and nine smaller Ministral 3 models https://mistral.ai/news/mistral-3 Sophia Yang, Ph.D. @sophiamyang Excited to introduce @MistralAI Ministral 3 (3B, 8B, and 14B) and Mistral Large 3 (sparse MOE trained w/ 41B active and 675B total parameters) - All open weights under the Apache 2.0 license! Details in : 430 12:18 AM • Dec 3, 2025 Simon Willison @simonw Five new Mistral models today, but the one I'm most excited about is this tiny 3B model that has vision support and can run entirely in a browser! Xenova @xenovacom NEW: @MistralAI releases Mistral 3, a family of multimodal models, including three start-of-the-art dense models (3B, 8B, and 14B) and Mistral Large 3 (675B, 41B active). All Apache 2.0! Surprisingly, the 3B is small enough to run 100% locally in your browser on WebGPU! 1,339 12:28 AM • Dec 3, 2025 3Chinese short-video company Kuaishou launches Kling Video O1, saying it is the first multimodal AI model to unify video generation, editing, and post-production https://the-decoder.com/klings-video-o1-launches-as-the-first-all-in-one-video-model-for-generation-and-editing/ Kling AI @Kling_ai Kling Omni Launch Week Day 1: Introducing Kling O1 — Brand-New Creative Engine for Endless Possibilities! Input anything. Understand everything. Generate any vision. With true multimodal understanding, Kling O1 unifies your input across texts, images, and videos — making 2,513 10:55 PM • Dec 1, 2025 Kling AI @Kling_ai With Kling O1, you can be ANYWHERE you want! Whether it's a change in background, environment, or even a weather, simply have Kling O1 put you in your desired world seamlessly. 20 11:49 AM • Dec 2, 2025 Thursday 4th December 2025Friday 5th December 20251I wanted to take this opportunity to remind members that I also curate some content via public Notion pages * Web3 Resource Hub, * AI/ML Resource Hub, * Videos of conversations I've been privileged to host with web3/AI industry luminaries Links to these are consolidated below. I invite you to read the content and join the communities if they interest you https://linktr.ee/goolamabbas2If you use Telegram for work and for fun/learning, I created the Notion page below where I describe some methods I use to keep my Telegram chats organised and even though I don't travel to as many events, I also use this for keeping conversations organised during events/conferences https://yusuf-goolamabbas-53.notion.site/Some-tips-on-Telegram-conversation-organisation-and-getting-organised-prior-to-a-conference-18883c2ae30380f582d8ebce60d23850Saturday 6th December 20251Google unveils Titans, an architecture combining RNN speed with transformer accuracy for real-time learning, which can scale to context windows of 2M+ tokens Jeff Dean @JeffDean We have a long history of model architecture innovations, and we're continuing to explore new ideas. Here's some early work that shows promising results on extending the context length over which a model can reason over. Google Research @GoogleResearch Today at #NeurIPS2025, we present Titans, a new architecture that combines the speed of RNNs with the performance of Transformers. It uses deep neural memory to learn in real-time, effectively scaling to contexts larger than 2 million tokens. More at: goo.gle/3Kd5ojF 1,683 5:30 AM • Dec 5, 2025 https://the-decoder.com/google-outlines-miras-and-titans-a-possible-path-toward-continuously-learning-ai/2Hugging Face details how it used its new tool, Skills, to fine tune an LLM using Claude, including for writing scripts, submitting jobs to cloud GPUs, and more clem 🤗 @ClementDelangue We managed to get Claude code, Codex and Gemini CLI to train good AI models thanks to @huggingface skills and you can too even (especially?) if you've never trained a model before After changing the way we build software, AI might start to change the way we build AI 816 7:09 AM • Dec 5, 2025 ℏεsam @Hesamation Hugging Face just made fine-tuning 10x easier. they released HF skills that you can plug into Claude Code, Codex, and Gemini to: > write training scripts > submit jobs to cloud GPUs > monitor the progress > push the models to HF hubs it's not just for fine-tuning, but also model 480 1:55 AM • Dec 5, 2025 Readers who think others in their family, friends and acquaintances who have a curiosity in knowing more about rapidly evolving AI tools/services/use cases would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe The cover image of this newsletter via generated via the Flux.2 model within the Freepik tool via the following promptan ornate Baroque cathedral with gilded details and delicate frescoes, enveloped in the soft, ethereal mist of early morning, framed by a low-angle perspective that emphasizes its towering arches and majestic dome against a pastel skyRegards, Yusuf

## Publication Information

- [This Week in All Things AI](https://paragraph.com/@twiata/): Publication homepage
- [All Posts](https://paragraph.com/@twiata/): More posts from this publication
- [RSS Feed](https://api.paragraph.com/blogs/rss/@twiata): Subscribe to updates
- [Twitter](https://twitter.com/yusufg): Follow on Twitter