# forck.live — 22–28 December 2025 > AI lab announcements, news reports and podcasts. forck.live follows what companies announce about their own products. Press may corroborate an official story; it never leads and never appears as an item of its own. A missing date stays missing. A title is the publisher's words — where a company does not publish in English, an English rendering is listed separately as `Title in English` and named as this site's translation, never as theirs. This file is one issue of the weekly archive: every confirmed first-party announcement of ISO week 2025-W52, 22–28 December 2025, newest first. The window never moves, so this URL always names the same seven days — but a post published inside them and detected later is added when it arrives, so a closed week can still gain entries. Every entry is one first-party announcement, and carries the same eight fields in this order before any others: Company · Announced · Category · Coverage · Announcement · Group · Source · Record `Source` is the company's own page the announcement was traced to and `Record` is this site's page for it. Both are safe to cite. `Coverage` is a number of outlets, or `not counted` — which means this record reached no count of its own, never that nobody wrote about it. `Category` is this tracker's own analysis, and is authoritative wherever it is stated. `not stated` means either that nothing has analysed the post yet or that the analysis placed it outside this taxonomy; neither is a verdict about the announcement. `Announcement` is whether the post is news of something the company did, rather than a how-to, a customer story, a changelog listing or a staffing note. `Group` is which of this record's groups the entry falls in, carried on the entry itself so it holds where the groups are headings and where they are not. One of: `models`, `covered`, `announcements`, `routine`. Anything after `Record` is held detail that only some announcements carry. - This week's page: https://forck.live/week/2025-W52 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Vercel — Stopping the slow death of internal tools - Company: Vercel (vercel.com) - Announced: 2025-12-27T13:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/stopping-the-slow-death-of-internal-tools - Record: https://forck.live/items/20770-stopping-the-slow-death-of-internal-tools - Subject: v0 / AI SDK / AI Gateway Vercel's v0 AI agent now enables business users to build and publish internal tools with real code, integrating with existing systems and SSO, while security is enforced at the platform level rather than relying on individual implementations. ### LG AI Research — [AI Ethics Seminar 2025 EP.3] Agentic AI Threat Modeling and Guardrail Implementation Strategy - Company: LG AI Research (lgresearch.ai) - Announced: 2025-12-24 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=609 - Record: https://forck.live/items/4767-ai-ethics-seminar-2025-ep-3-agentic-ai-threat-modeling-and-guardrail - Subject: EXAONE LG AI Research's AI Ethics Seminar discusses threat modeling for Agentic AI, highlighting new security challenges such as indirect prompt injection and intent breaking, and emphasizing the shift from content safety to action safety and goal integrity. ### Google DeepMind — Google's year in review: 8 areas with research breakthroughs in 2025 - Company: Google DeepMind (deepmind.google) - Announced: 2025-12-23T17:01:02+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://deepmind.google/blog/googles-year-in-review-8-areas-with-research-breakthroughs-in-2025/ - Record: https://forck.live/items/1259-google-s-year-in-review-8-areas-with-research-breakthroughs-in-2025 - Subject: Gemini The source text is a title only, stating that Google's 2025 recap covers research breakthroughs of the year. No specific product, model, or announcement details are provided. ### Hugging Face — AprielGuard: A Guardrail for Safety and Adversarial Robustness in Modern LLM Systems - Company: Hugging Face (huggingface.co) - Announced: 2025-12-23T14:07:35+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/ServiceNow-AI/aprielguard - Record: https://forck.live/items/1568-aprielguard-a-guardrail-for-safety-and-adversarial-robustness-in-modern-llm - Subject: Platform ServiceNow AI introduced AprielGuard, an 8B parameter safety–security safeguard model designed to detect 16 categories of safety risks and a wide range of adversarial attacks, available in reasoning and non-reasoning modes. ### Vercel — Pixel Portraits: AI generated trading cards - Company: Vercel (vercel.com) - Announced: 2025-12-23T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/pixel-portraits-ai-generated-trading-cards - Record: https://forck.live/items/20771-pixel-portraits-ai-generated-trading-cards - Subject: v0 / AI SDK / AI Gateway Vercel describes the creation of an AI-powered photobooth that generated pixel-art trading cards for attendees of Next.js Conf and Ship AI events. The company also released a v0 template that brings the same concept into the browser, using the Vercel AI Gateway and Fal integration. ### Alibaba — Qwen-Image-Edit-2511: Improve Consistency - Company: Alibaba (alibaba.com) - Announced: 2025-12-23T05:08:30+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwen.ai/blog?id=qwen-image-edit-2511 - Record: https://forck.live/items/4588-qwen-image-edit-2511-improve-consistency - Subject: Qwen Alibaba released Qwen-Image-Edit-2511, an updated image editing model with improved character consistency, multi-person consistency, built-in LoRA integration, and enhanced geometric reasoning. ### Kyutai — CASA: The return of the cross-attention - Company: Kyutai (kyutai.org) - Announced: 2025-12-23 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://kyutai.org/casa - Record: https://forck.live/items/18350-casa-the-return-of-the-cross-attention - Subject: Moshi / Pocket TTS Kyutai's research revisits cross-attention (CA) as a fusion mechanism for vision-language models (VLMs), finding that when trained under comparable conditions, CA-based models achieve performance close to token insertion models on most benchmarks, with a remaining gap on some high-resolution tasks. The paper highlights CA's favorable compute and memory scaling for streaming applications, such as live video captioning, where memory and inference costs remain nearly constant as more images are processed. ### Baidu — Best Text model from China in LMArena is now ERNIE-5.0-Preview-1203! - Company: Baidu (baidu.com) - Announced: 2025-12-23 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://ernie.baidu.com/blog/posts/ernie-5.0-preview-1203-release-on-lmarena/ - Record: https://forck.live/items/4647-best-text-model-from-china-in-lmarena-is-now-ernie-5-0-preview-1203 - Subject: ERNIE Baidu announces that its ERNIE-5.0-Preview-1203 model has achieved a score of 1,451 points on LMArena, leading in Creative Writing and Hard Prompts. The model is now available for developers to experience on LMArena. ### MiniMax — MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks - Company: MiniMax (minimax.io) - Announced: 2025-12-23 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.minimax.io/blog/minimax-m21 - Record: https://forck.live/items/3403-minimax-m2-1-significantly-enhanced-multi-language-programming-built-for-real - Subject: MiniMax models MiniMax M2.1 is an enhanced model with industry-leading multi-language programming capabilities across Rust, Java, Go, C++, TypeScript and more, featuring enhanced mobile development, aesthetic web design, concise responses, and robust Agent framework generalization. ### Together AI — MiniMax Speech 2.6 Turbo now available natively on Together AI - Company: Together AI (together.ai) - Announced: 2025-12-23 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.together.ai/blog/minimax-speech-2-6 - Record: https://forck.live/items/2391-minimax-speech-2-6-turbo-now-available-natively-on-together-ai - Subject: Inference platform Together AI announces that MiniMax Speech 2.6 Turbo is now available on its dedicated infrastructure, offering sub-250ms latency, 40+ languages with streaming inline switching, 10-second voice cloning, automatic emotional awareness, and top ranking on the Artificial Analysis Arena. It is the only platform providing this model on dedicated infrastructure alongside LLM and STT workloads. ### Replit — Now You Can Build with Replit in ChatGPT - Company: Replit (replit.com) - Announced: 2025-12-22T21:49:55+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/replit-in-chatgpt - Record: https://forck.live/items/17922-now-you-can-build-with-replit-in-chatgpt - Subject: Replit Agent Replit has integrated with ChatGPT, allowing users to create, update, and deploy Replit Apps directly within ChatGPT conversations by tagging @replit and describing the desired app. The integration works in markets where the ChatGPT App Store is available and requires a ChatGPT account (Free, Plus, Pro, or Teams) and a Replit account (free Starter works, Core or Teams recommended for private apps and databases). Each conversation can work with one Replit App, and work done through this integration uses Replit Agent credits in the user's Replit account. ### Alibaba — Qwen3-TTS Steps Up: Voice Cloning and Voice Design! - Company: Alibaba (alibaba.com) - Announced: 2025-12-22T16:00:45+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwen.ai/blog?id=qwen3-tts-vc-voicedesign - Record: https://forck.live/items/4591-qwen3-tts-steps-up-voice-cloning-and-voice-design - Subject: Qwen Alibaba's Qwen3-TTS family launches two new models: a voice design model (Qwen3-TTS-VD-Flash) and a voice cloning model (Qwen3-TTS-VC-Flash), both accessible via the Qwen API. The voice design model supports complex natural language instructions for fine-grained control over timbre, prosody, emotion, and persona, and the voice cloning model supports 3-second voice cloning in 10 languages. The models claim superior performance on benchmarks such as InstructTTS-Eval and MiniMax TTS Multilingual Test Set. ### Harvey — Harvey In Practice: How to Evaluate Your Arguments - Company: Harvey (harvey.ai) - Announced: 2025-12-22T16:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/harvey-in-practice-evaluate-your-arguments - Record: https://forck.live/items/17632-harvey-in-practice-how-to-evaluate-your-arguments - Subject: Harvey Harvey published a guide showing how its AI can be used to review and strengthen complex legal filings like class action complaints. The tool analyzes draft complaints, highlights potential weaknesses, and provides citation-backed suggestions for improvement. ### Vercel — AI SDK 6 - Company: Vercel (vercel.com) - Announced: 2025-12-22T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/ai-sdk-6 - Record: https://forck.live/items/20773-ai-sdk-6 - Subject: v0 / AI SDK / AI Gateway Introducing agents, tool execution approval, DevTools, full MCP support, reranking, image editing, and more. ### Vercel — We removed 80% of our agent’s tools - Company: Vercel (vercel.com) - Announced: 2025-12-22T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/we-removed-80-percent-of-our-agents-tools - Record: https://forck.live/items/20772-we-removed-80-of-our-agent-s-tools - Subject: v0 / AI SDK / AI Gateway Vercel rebuilt its internal text-to-SQL agent d0 by replacing 16 specialized tools with a single tool that executes arbitrary bash commands, giving Claude Opus 4.5 direct access to the file system. The new agent achieved a 100% success rate (up from 80%), 3.5x faster execution, 37% fewer tokens, and 42% fewer steps across five benchmark queries. ### Harvey — Centralize and Customize Legal Tools With MCP - Company: Harvey (harvey.ai) - Announced: 2025-12-22T13:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/harvey-mcp-overview - Record: https://forck.live/items/17633-centralize-and-customize-legal-tools-with-mcp - Subject: Harvey Harvey is adopting the Model Context Protocol (MCP) to allow customers to integrate their own tools and third-party systems more flexibly. Harvey can operate as an MCP client to orchestrate external services, or as an MCP server to expose its legal reasoning capabilities to other systems, while maintaining centralized permissions and auditing. ### Z.ai — GLM-4.7 - Company: Z.ai (z.ai) - Announced: 2025-12-22 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://docs.z.ai/release-notes/new-released#2025-12-22 - Record: https://forck.live/items/4455-glm-4-7 - Subject: GLM Z.ai released GLM-4.7, an updated foundation model with improvements in coding, reasoning, and agentic capabilities, including open-source SOTA performance on benchmarks and enhanced agentic coding. ### xAI — Supporting the DOW's mission with AI - Company: xAI (x.ai) - Announced: 2025-12-22 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://x.ai/news/us-gov-dept-of-war - Record: https://forck.live/items/3479-supporting-the-dow-s-mission-with-ai - Subject: Grok xAI announces it has been selected by the US Department of War to deliver Frontier AI. ### xAI — Grok Collections API - Company: xAI (x.ai) - Announced: 2025-12-22 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://x.ai/news/grok-collections-api - Record: https://forck.live/items/3478-grok-collections-api - Subject: Grok xAI announced the Grok Collections API, described as a state-of-the-art RAG system built directly into their API. ### OpenAI — One in a million: celebrating the customers shaping AI’s future - Company: OpenAI (openai.com) - Announced: 2025-12-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/one-in-a-million-customers - Record: https://forck.live/items/323-one-in-a-million-celebrating-the-customers-shaping-ai-s-future - Subject: GPT / ChatGPT / API OpenAI celebrates reaching more than one million customers and highlights companies using its AI to transform work. ### OpenAI — Continuously hardening ChatGPT Atlas against prompt injection - Company: OpenAI (openai.com) - Announced: 2025-12-22 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/hardening-atlas-against-prompt-injection - Record: https://forck.live/items/322-continuously-hardening-chatgpt-atlas-against-prompt-injection - Subject: GPT / ChatGPT / API OpenAI is strengthening ChatGPT Atlas against prompt injection attacks using automated red teaming trained with reinforcement learning.