# forck.live — 15–21 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-W51, 15–21 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-W51
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Odyssey — The Dawn of a World Simulator

- Company: Odyssey (odyssey.systems)
- Announced: 2025-12-20
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://odyssey.systems/the-dawn-of-a-world-simulator
- Record: https://forck.live/items/18603-the-dawn-of-a-world-simulator
- Subject: Odyssey world models

Odyssey describes Odyssey-2 Pro, a causal world model that learns to predict the world's evolution from trillions of video and interaction observations. The model aims to internalize physics, human behavior, and cause-and-effect without hand-crafted rules, potentially serving as a general simulator for domains such as robotics, science, and gaming.

### Vercel — Our $1 million hacker challenge for React2Shell

- Company: Vercel (vercel.com)
- Announced: 2025-12-19T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/our-million-dollar-hacker-challenge-for-react2shell
- Record: https://forck.live/items/20774-our-1-million-hacker-challenge-for-react2shell
- Subject: v0 / AI SDK / AI Gateway

In the weeks following React2Shell's disclosure, our firewall blocked over 6 million exploit attempts targeting deployments running vulnerable versions of Next.js, with 2.3 million in a single 24-hour period at peak.

### Alibaba — Qwen-Image-Layered: Layered Decomposition for Inherent Editablity

- Company: Alibaba (alibaba.com)
- Announced: 2025-12-19T05:08:30+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen-image-layered
- Record: https://forck.live/items/4589-qwen-image-layered-layered-decomposition-for-inherent-editablity
- Subject: Qwen

Alibaba introduces Qwen-Image-Layered, a model that decomposes images into multiple RGBA layers for independent editing, supporting variable-layer and recursive decomposition.

### Supabase — Sync Stripe Data to Your Supabase Database in One Click

- Company: Supabase (supabase.com)
- Announced: 2025-12-19
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/stripe-sync-engine-integration
- Record: https://forck.live/items/21311-sync-stripe-data-to-your-supabase-database-in-one-click
- Subject: Supabase

Supabase announced a partnership with Stripe and official support for the Stripe Sync Engine in the Supabase Dashboard, providing a one-click integration that syncs Stripe data directly into a Supabase database. The integration uses webhooks for real-time updates and scheduled backfills for historical data, allowing users to query customers, subscriptions, invoices, and payments with standard SQL. Stripe engineers contributed improvements to the open-source sync engine, including incremental sync, flexible JSONB storage, and a new CLI.

### Google Research — Google Research 2025: Bolder breakthroughs, bigger impact

- Company: Google Research (research.google)
- Announced: 2025-12-18T23:29:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://research.google/blog/google-research-2025-bolder-breakthroughs-bigger-impact/
- Record: https://forck.live/items/1381-google-research-2025-bolder-breakthroughs-bigger-impact
- Subject: Research

This is a year-end review post by Google Research summarizing their 2025 achievements across AI, science, and society, without announcing new products or model releases.

### Harvey — Deliver on Quality With Knowledge and Workflow Enhancements

- Company: Harvey (harvey.ai)
- Announced: 2025-12-18T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/knowledge-workflows-enhancements
- Record: https://forck.live/items/17634-deliver-on-quality-with-knowledge-and-workflow-enhancements
- Subject: Harvey

Harvey expanded its platform with over 70 new local legal data sources, bringing the total to 200, and added multi-source querying, optional workflow steps, and an automatic workflow optimization feature. The updates allow lawyers to query Ask LexisNexis AI alongside document uploads and attach regional knowledge sources, web search, and Vault files directly into custom workflows.

### Lovable — Lovable raises $330M to power the age of the builder

- Company: Lovable (lovable.dev)
- Announced: 2025-12-18T13:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/series-b
- Record: https://forck.live/items/12113-lovable-raises-330m-to-power-the-age-of-the-builder
- Subject: Lovable / AI app builder

Lovable raised $330 million in Series B funding at a $6.6 billion valuation, led by CapitalG and Menlo Ventures' Anthology fund, with participation from venture arms of NVIDIA, Salesforce, Databricks, Deutsche Telekom, Atlassian, and HubSpot. The company reported 100,000+ new projects built daily, 25 million+ total projects created in its first year, and 6 million+ daily visits to Lovable-built sites and apps.

### OpenAI — Evaluating chain-of-thought monitorability

- Company: OpenAI (openai.com)
- Announced: 2025-12-18T12:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/evaluating-chain-of-thought-monitorability
- Record: https://forck.live/items/324-evaluating-chain-of-thought-monitorability
- Subject: GPT / ChatGPT / API

OpenAI announced a new framework and evaluation suite for chain-of-thought monitorability, with 13 evaluations across 24 environments, showing that monitoring internal reasoning is more effective than monitoring outputs.

### OpenAI — Updating our Model Spec with teen protections

- Company: OpenAI (openai.com)
- Announced: 2025-12-18T11:00:00+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/updating-model-spec-with-teen-protections
- Record: https://forck.live/items/327-updating-our-model-spec-with-teen-protections
- Subject: GPT / ChatGPT / API

OpenAI updated its Model Spec to include Under-18 Principles for safer, age-appropriate ChatGPT interactions with teens.

### OpenAI — AI literacy resources for teens and parents

- Company: OpenAI (openai.com)
- Announced: 2025-12-18T11:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/ai-literacy-resources-for-teens-and-parents
- Record: https://forck.live/items/326-ai-literacy-resources-for-teens-and-parents
- Subject: GPT / ChatGPT / API

OpenAI published new AI literacy resources aimed at teens and parents, covering responsible use, critical thinking, healthy boundaries, and handling emotional topics.

### OpenAI — Deepening our collaboration with the U.S. Department of Energy

- Company: OpenAI (openai.com)
- Announced: 2025-12-18T11:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/us-department-of-energy-collaboration
- Record: https://forck.live/items/325-deepening-our-collaboration-with-the-u-s-department-of-energy
- Subject: GPT / ChatGPT / API

OpenAI and the U.S. Department of Energy signed a memorandum of understanding to deepen collaboration on AI and advanced computing to support scientific discovery, building on ongoing work with national laboratories.

### Replit — Inside Replit’s Snapshot Engine: The Tech Making AI Agents Safe

- Company: Replit (replit.com)
- Announced: 2025-12-18T01:29:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/inside-replits-snapshot-engine
- Record: https://forck.live/items/17926-inside-replit-s-snapshot-engine-the-tech-making-ai-agents-safe
- Subject: Replit Agent

Replit describes how its snapshot engine, built on Copy-on-Write block devices and Git versioning, allows users to revert code and database changes made by its AI Agent. The system uses separate development and production databases, restricting the Agent to the development database, and provides point-in-time recovery of filesystems and databases.

### Luma — Introducing Ray3 Modify

- Company: Luma (lumalabs.ai)
- Announced: 2025-12-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lumalabs.ai/news/ray3-modify
- Record: https://forck.live/items/7974-introducing-ray3-modify
- Subject: Ray / Dream Machine

Luma introduced Ray3 Modify, a new feature in Dream Machine that enables video editing and reimagination with keyframe and character reference controls. The update includes improved Modify Video capabilities for scene-aware transformations, as well as new Modify with Keyframes and Character Reference features.

### Amp — A Codebase by an Agent for an Agent

- Company: Amp (ampcode.com)
- Announced: 2025-12-18
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ampcode.com/by-an-agent-for-an-agent
- Record: https://forck.live/items/7771-a-codebase-by-an-agent-for-an-agent
- Subject: Amp

The Amp agent autonomously wrote a TUI framework, and the author observed that allowing the agent to make its own naming and structure decisions improved its performance.

### Amp — Agentic Review

- Company: Amp (ampcode.com)
- Announced: 2025-12-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/agentic-code-review
- Record: https://forck.live/items/7770-agentic-review
- Subject: Amp

Amp introduced a dedicated agent for code review within its VS Code extension. The agent uses Gemini 3 Pro to provide deeper analysis and actionable feedback compared to a previous single-shot LLM version.

### Together AI — Rime voice models now available on Together AI

- Company: Together AI (together.ai)
- Announced: 2025-12-18
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.together.ai/blog/rime-voice-models-now-available-on-together-ai
- Record: https://forck.live/items/2392-rime-voice-models-now-available-on-together-ai
- Subject: Inference platform

Together AI adds Rime Arcana v2 and Mist v2 voice models to its Model Library, offering expressive conversational voices with 40+ voices and deterministic pronunciation control for enterprise voice applications.

### Hugging Face — Tokenization in Transformers v5: Simpler, Clearer, and More Modular

- Company: Hugging Face (huggingface.co)
- Announced: 2025-12-18
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/tokenizers
- Record: https://forck.live/items/1569-tokenization-in-transformers-v5-simpler-clearer-and-more-modular
- Subject: Platform

Hugging Face announces Transformers v5, a major redesign of tokenization that separates tokenizer architecture from trained vocabulary, making tokenizers inspectable, customizable, and trainable with less friction.

### OpenAI — Introducing GPT-5.2-Codex

- Company: OpenAI (openai.com)
- Announced: 2025-12-18
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/introducing-gpt-5-2-codex
- Record: https://forck.live/items/328-introducing-gpt-5-2-codex
- Subject: GPT / ChatGPT / API

OpenAI announces GPT-5.2-Codex, a coding model with long-horizon reasoning, large-scale code transformations, and enhanced cybersecurity capabilities.

### Google Antigravity — Gemini 3 Flash in Google Antigravity, agentic coding at the speed of thought

- Company: Google Antigravity (antigravity.google)
- Announced: 2025-12-17T18:19:47+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://antigravity.google/blog/gemini-3-flash-in-google-antigravity
- Record: https://forck.live/items/4558-gemini-3-flash-in-google-antigravity-agentic-coding-at-the-speed-of-thought
- Subject: Antigravity

Google Antigravity announces the preview availability of Gemini 3 Flash in its development environment, providing faster agentic coding capabilities.

### Harvey — Harvey In Practice: How to Analyze Multiple Lease Agreements

- Company: Harvey (harvey.ai)
- Announced: 2025-12-17T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-in-practice-analyze-multiple-lease-agreements
- Record: https://forck.live/items/17635-harvey-in-practice-how-to-analyze-multiple-lease-agreements
- Subject: Harvey

Harvey published a video in its 'Harvey In Practice' series demonstrating how legal teams can upload multiple lease agreements and generate structured summaries of key terms such as base rent, extension options, and assignment rights. The video shows the workflow of extracting and comparing terms across agreements, with each extracted term linked to the source section of the lease.

### NVIDIA — The Open Evaluation Standard: Benchmarking NVIDIA Nemotron 3 Nano with NeMo Evaluator

- Company: NVIDIA (nvidia.com)
- Announced: 2025-12-17T13:22:18+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/nvidia/nemotron-3-nano-evaluation-recipe
- Record: https://forck.live/items/1570-the-open-evaluation-standard-benchmarking-nvidia-nemotron-3-nano-with-nemo

NVIDIA released the Nemotron 3 Nano 30B A3B model with an open evaluation approach, publishing the complete evaluation recipe using the NeMo Evaluator library for reproducibility.

### Google DeepMind — Gemini 3 Flash: frontier intelligence built for speed

- Company: Google DeepMind (deepmind.google)
- Announced: 2025-12-17T11:58:17+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/gemini-3-flash-frontier-intelligence-built-for-speed/
- Record: https://forck.live/items/1260-gemini-3-flash-frontier-intelligence-built-for-speed
- Subject: Gemini

Gemini 3 Flash is a model that offers frontier intelligence built for speed at a fraction of the cost.

### OpenAI — Introducing OpenAI Academy for News Organizations

- Company: OpenAI (openai.com)
- Announced: 2025-12-17T06:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-academy-for-news-organizations
- Record: https://forck.live/items/330-introducing-openai-academy-for-news-organizations
- Subject: GPT / ChatGPT / API

OpenAI launches the OpenAI Academy for News Organizations, a learning hub to help newsrooms use AI effectively, with training, use cases, and guidance.

### Ant Ling — The Community Stories of vLLM and SGLang

- Company: Ant Ling (ant-ling.com)
- Announced: 2025-12-17
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://developer.ant-ling.com/en/blogs/llm-landscape-vllm-sgl
- Record: https://forck.live/items/22334-the-community-stories-of-vllm-and-sglang
- Subject: Ling / Ring / Ming

Ant Ling's article compares the open-source LLM inference engines vLLM and SGLang, noting that vLLM handled up to 5× the traffic and boosted throughput by as much as 30× compared with a Hugging Face Transformers–based backend. SGLang extends vLLM's PagedAttention with RadixAttention to preserve and reuse KV cache across requests. The article also details the projects' origins at UC Berkeley, their community metrics, and key milestones through 2025.

### Supabase — Building ChatGPT Apps with Supabase Edge Functions and mcp-use

- Company: Supabase (supabase.com)
- Announced: 2025-12-17
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/building-chatgpt-apps-with-supabase
- Record: https://forck.live/items/21312-building-chatgpt-apps-with-supabase-edge-functions-and-mcp-use
- Subject: Supabase

Supabase published a guide showing developers how to build ChatGPT apps that connect to a Supabase database using mcp-use, an open-source SDK for deploying MCP servers on Supabase Edge Functions. The guide covers creating an MCP server and web components that let ChatGPT users explore database schemas, view table data, and run SQL queries through interactive widgets.

### Upstage — Introducing Document Parse: Enhanced mode

- Company: Upstage (upstage.ai)
- Announced: 2025-12-17
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/blog/en/document-parse-enhanced
- Record: https://forck.live/items/18457-introducing-document-parse-enhanced-mode
- Subject: Solar Pro / Solar Mini

Upstage introduced Document Parse Enhanced, a new processing mode that uses vision language models to extract complex tables, charts, checkboxes, and diagrams from documents. It achieves higher accuracy on these elements compared to standard parsing while maintaining practical latency. The mode is available as part of Document Parse, alongside standard and auto modes.

### Upstage — 2025 in Review: South Korea’s Leading AI Innovator Marks Breakout Year After U.S. Launch

- Company: Upstage (upstage.ai)
- Announced: 2025-12-17
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/news/2025-in-review-south-koreas-leading-ai-innovator-marks-breakout-year-after-u-s-launch
- Record: https://forck.live/items/18425-2025-in-review-south-korea-s-leading-ai-innovator-marks-breakout-year-after-u
- Subject: Solar Pro / Solar Mini

Upstage AI announced a $45 million Series B bridge round, bringing total funding to $157 million, with Amazon Web Services and AMD becoming minority investors in August. The company reports 95%+ accuracy on complex document types and serves Fortune 500 companies including Samsung, processing millions of pages daily. Upstage was named to CB Insights' AI 100, InsurTech 50, and FinTech 100 lists following its U.S. market debut.

### Amp — Gemini 3 Flash in the Search Subagent

- Company: Amp (ampcode.com)
- Announced: 2025-12-17
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/gemini-3-flash-search
- Record: https://forck.live/items/7772-gemini-3-flash-in-the-search-subagent
- Subject: Amp

Amp's codebase search subagent now uses Gemini 3 Flash instead of Haiku 4.5, resulting in 3x faster performance with the same quality, better parallel tool calls, and earlier conclusion.

### xAI — Grok Voice Agent API

- Company: xAI (x.ai)
- Announced: 2025-12-17
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/grok-voice-agent-api
- Record: https://forck.live/items/3480-grok-voice-agent-api
- Subject: Grok

xAI announced the Grok Voice Agent API, making Grok Voice available to developers.

### Mistral AI — Introducing Mistral OCR 3

- Company: Mistral AI (mistral.ai)
- Announced: 2025-12-17
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-ocr-3
- Record: https://forck.live/items/3307-introducing-mistral-ocr-3
- Subject: Mistral / Le Chat

Mistral AI announced Mistral OCR 3, a research paper.

### Together AI — Research POV: Yes, AGI Can Happen – A Computational Perspective

- Company: Together AI (together.ai)
- Announced: 2025-12-17
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/research-pov-yes-agi-can-happen
- Record: https://forck.live/items/2393-research-pov-yes-agi-can-happen-a-computational-perspective
- Subject: Inference platform

Together AI's VP of Kernels published a post arguing that AI is not hitting a hardware wall and that better software-hardware co-design can unlock significant performance gains, supporting the possibility of AGI.

### OpenAI — Developers can now submit apps to ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2025-12-17
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/developers-can-now-submit-apps-to-chatgpt
- Record: https://forck.live/items/332-developers-can-now-submit-apps-to-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI announced that developers can now submit apps for review and publication in ChatGPT, with approved apps appearing in a new in-product directory. Updated tools, guidelines, and the Apps SDK are provided to help build chat-native experiences.

### OpenAI — The state of enterprise AI

- Company: OpenAI (openai.com)
- Announced: 2025-12-17
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/business/guides-and-resources/the-state-of-enterprise-ai-2025-report
- Record: https://forck.live/items/331-the-state-of-enterprise-ai
- Subject: GPT / ChatGPT / API

A data-driven report on enterprise AI adoption, showing how organizations move from experimentation to real productivity gains and new capabilities.

### Lovable — Perplexity, ElevenLabs, Firecrawl, and Miro: now in Lovable

- Company: Lovable (lovable.dev)
- Announced: 2025-12-16T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/ai-connectors
- Record: https://forck.live/items/12114-perplexity-elevenlabs-firecrawl-and-miro-now-in-lovable
- Subject: Lovable / AI app builder

Lovable has launched connectors for Perplexity, ElevenLabs, Firecrawl, and Miro, enabling users to integrate AI search, voice and sound generation, web scraping, and visual workspace capabilities into apps built on the platform. Firecrawl is free for Lovable users through January 2026 when opting for a managed instance.

### Vercel — Cline now runs on Vercel AI Gateway

- Company: Vercel (vercel.com)
- Announced: 2025-12-16T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/cline-on-ai-gateway
- Record: https://forck.live/items/20775-cline-now-runs-on-vercel-ai-gateway
- Subject: v0 / AI SDK / AI Gateway

Cline, an open-source coding agent, now routes requests through Vercel AI Gateway. In A/B testing, P99 streaming latency improved by 10–14% across its most-used models and API error rates dropped by 43.8%.

### Google DeepMind — Gemma Scope 2: helping the AI safety community deepen understanding of complex language model behavior

- Company: Google DeepMind (deepmind.google)
- Announced: 2025-12-16T10:14:24+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://deepmind.google/blog/gemma-scope-2-helping-the-ai-safety-community-deepen-understanding-of-complex-language-model-behavior/
- Record: https://forck.live/items/1261-gemma-scope-2-helping-the-ai-safety-community-deepen-understanding-of-complex
- Subject: Gemini

Gemma Scope 2, an open interpretability tool, is now available for the entire Gemma 3 family of language models.

### OpenAI — Evaluating AI’s ability to perform scientific research tasks

- Company: OpenAI (openai.com)
- Announced: 2025-12-16T09:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/frontierscience
- Record: https://forck.live/items/333-evaluating-ai-s-ability-to-perform-scientific-research-tasks
- Subject: GPT / ChatGPT / API

OpenAI introduces FrontierScience, a benchmark for testing AI reasoning in physics, chemistry, and biology to measure progress toward real scientific research.

### OpenAI — Measuring AI’s capability to accelerate biological research

- Company: OpenAI (openai.com)
- Announced: 2025-12-16T08:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/accelerating-biological-research-in-the-wet-lab
- Record: https://forck.live/items/334-measuring-ai-s-capability-to-accelerate-biological-research
- Subject: GPT / ChatGPT / API

OpenAI introduces a real-world evaluation framework to measure AI's capability to accelerate biological research, using GPT-5 to optimize a molecular cloning protocol and exploring promise and risks.

### Supabase — Own Your Observability: Supabase Metrics API

- Company: Supabase (supabase.com)
- Announced: 2025-12-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/metrics-api-observability
- Record: https://forck.live/items/21313-own-your-observability-supabase-metrics-api
- Subject: Supabase

Supabase published enhanced documentation for its Metrics API, which exposes roughly 200 Postgres performance and health metrics in Prometheus exposition format. The API allows users to stream database telemetry into any Prometheus-compatible observability stack, including Grafana Cloud, Datadog, and AWS Managed Prometheus. The Metrics API is available on all hosted Supabase projects.

### Factory AI — Evaluating Context Compression for AI Agents

- Company: Factory AI (factory.com)
- Announced: 2025-12-16
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/evaluating-compression
- Record: https://forck.live/items/17833-evaluating-context-compression-for-ai-agents
- Subject: Droid

Factory Research built an evaluation framework to measure how well different context compression strategies preserve information for long-running AI agent sessions. They tested three approaches—Factory's anchored iterative summarization, OpenAI's /responses/compact endpoint, and Anthropic's built-in compression—and found that structured summarization retains more useful information than the alternatives without sacrificing compression efficiency.

### Amp — Clickable Diagrams

- Company: Amp (ampcode.com)
- Announced: 2025-12-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/clickable-diagrams
- Record: https://forck.live/items/7773-clickable-diagrams
- Subject: Amp

Amp's editor extensions now generate Mermaid diagrams with clickable code references, allowing users to jump to the relevant file or code snippet by clicking on a node or edge.

### LG AI Research — AI Scientist & AI Victor: Two Geniuses Discuss 'The Future of AI' - Part 2

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-12-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=607
- Record: https://forck.live/items/4768-ai-scientist-ai-victor-two-geniuses-discuss-the-future-of-ai-part-2
- Subject: EXAONE

This is a discussion between Honglak Lee (CSAI at LG AI Research) and Saedol Lee about the impact of AI on education, critical thinking, and the ideal human talent in the AI era. No new product, model, or capability is announced.

### OpenAI — Staying ahead in the age of AI

- Company: OpenAI (openai.com)
- Announced: 2025-12-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/business/guides-and-resources/staying-ahead-in-the-age-of-ai
- Record: https://forck.live/items/336-staying-ahead-in-the-age-of-ai
- Subject: GPT / ChatGPT / API

OpenAI provides a guide for leaders on building AI-ready organizations through strategy, training, governance, and innovation.

### OpenAI — The new ChatGPT Images is here

- Company: OpenAI (openai.com)
- Announced: 2025-12-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/new-chatgpt-images-is-here
- Record: https://forck.live/items/335-the-new-chatgpt-images-is-here
- Subject: GPT / ChatGPT / API

OpenAI announced an upgraded image generation model powering ChatGPT Images, with faster generation and better edits, rolling out to all ChatGPT users and available in the API as GPT-Image-1.5.

### Replit — Enabling Agent 3 to Self-Test at Scale with REPL-Based Verification

- Company: Replit (replit.com)
- Announced: 2025-12-15T18:31:59+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/automated-self-testing
- Record: https://forck.live/items/17928-enabling-agent-3-to-self-test-at-scale-with-repl-based-verification
- Subject: Replit Agent

Replit describes how it built a REPL-based verification system combining code execution with browser automation to detect 'Potemkin interfaces'—features that appear functional but are not. The system enables Agent 3 to work autonomously for over 200 minutes by catching these failures early and preventing compounding mistakes.

### Google Research — Gemini-backed Paper Assistant Tool provides automated feedback for theoretical computer scientists at STOC 2026

- Company: Google Research (research.google)
- Announced: 2025-12-15T17:37:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/gemini-provides-automated-feedback-for-theoretical-computer-scientists-at-stoc-2026/
- Record: https://forck.live/items/1382-gemini-backed-paper-assistant-tool-provides-automated-feedback-for-theoretical
- Subject: Research

Google Research describes the Paper Assistant Tool (PAT), which uses Gemini 2.5 Deep Think to provide automated pre-submission feedback to authors at the STOC 2026 conference, helping identify errors and improve paper clarity.

### Hugging Face — CUGA on Hugging Face: Democratizing Configurable AI Agents

- Company: Hugging Face (huggingface.co)
- Announced: 2025-12-15T16:01:04+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/ibm-research/cuga-on-hugging-face
- Record: https://forck.live/items/1571-cuga-on-hugging-face-democratizing-configurable-ai-agents
- Subject: Platform

Hugging Face and IBM Research announce the open-source release of CUGA, a configurable generalist AI agent, under Apache 2.0 license. The post highlights CUGA's state-of-the-art benchmark performance, its architectural patterns, and its integration with Hugging Face Spaces and Langflow. The agent supports open models and runs on Groq's LPUs for fast inference.

### Vercel — Build smarter workflows with Notion and v0

- Company: Vercel (vercel.com)
- Announced: 2025-12-15T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/build-smarter-workflows-with-notion-and-v0
- Record: https://forck.live/items/20777-build-smarter-workflows-with-notion-and-v0
- Subject: v0 / AI SDK / AI Gateway

Vercel announced that v0 can now be securely connected to a user's Notion workspace, allowing it to read from and write to Notion databases and docs. The integration enables building dashboards, prototypes, and summaries grounded in existing team knowledge, and supports write access for v0 to contribute content back into Notion.

### Vercel — How to prompt v0

- Company: Vercel (vercel.com)
- Announced: 2025-12-15T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-to-prompt-v0
- Record: https://forck.live/items/20776-how-to-prompt-v0
- Subject: v0 / AI SDK / AI Gateway

Vercel published a guide on how to prompt its AI tool v0, presenting a three-part framework: product surface, context of use, and constraints & taste. The guide includes test results showing that more specific prompts lead to faster generation, fewer lines of code, and lower credit costs.

### Together AI — Announcing native availability of NVIDIA Nemotron 3 Nano, NVIDIA’s latest reasoning model

- Company: Together AI (together.ai)
- Announced: 2025-12-15
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/nemotron-3-nano-now-available-on-together-ai
- Record: https://forck.live/items/2394-announcing-native-availability-of-nvidia-nemotron-3-nano-nvidia-s-latest
- Subject: Inference platform

Together AI announces the availability of NVIDIA's Nemotron 3 Nano model on its platform, a reasoning model with a hybrid Mamba-Transformer architecture and open weights.