Frontier Labs
Date-range scope (new info surfaced): 2025-12-26 to 2025-12-30 (inclusive).
Acquisitions (frontier AI orgs)
- Meta → acquired Manus (AI agent startup; founded in China, headquartered in Singapore)
- Decision (announced): Meta said it would acquire Manus, continue operating/selling Manus as a standalone service, and integrate it into Meta products (including Meta AI). (reuters.com)
- Acquired company focus (what they do): Manus builds a general-purpose autonomous AI agent positioned as a “digital employee,” designed to execute tasks like research and automation with minimal prompting. (reuters.com)
- Approach / “how it works” (practical characteristics):
- Manus frames itself as an “execution layer” for autonomous agents—turning frontier model capabilities into end-to-end, operational task completion. (manus.im)
- Manus claims heavy operational scale since launch: 147T+ tokens processed and 80M+ virtual computers created (company-reported). (manus.im)
- Publicly marketed by doing dozens of tasks for users on X for free (distribution + product proof). (reuters.com)
- Competitive advantages / why this matters to Meta (inferred from disclosed facts):
- Faster path to monetization than “model-only” work: Manus reportedly had an annual revenue run rate of ~$125M earlier in 2025 and sells via subscription, which can give Meta a nearer-term commercial wedge for agents. (straitstimes.com)
- Agent differentiation: Manus positioned itself as a “leading” general-purpose agent and claimed it surpasses OpenAI DeepResearch (claim repeated by Meta/press; treat as unverified marketing until independently benchmarked). (reuters.com)
- Other unique deal details (useful context):
- Manus’ parent is Beijing Butterfly Effect Technology; it raised funding earlier in 2025 at ~$500M valuation in a round led by Benchmark (per reporting). (straitstimes.com)
- Manus says it will keep operating from Singapore and aims for no customer disruption (company statement). (manus.im)
Notable hires & talent moves (frontier model orgs)
- Meta Superintelligence Lab → continued “talent raid” from OpenAI (reported list)
- Decision pattern: Meta’s frontier effort is scaling via high-profile researcher hiring, including multiple departures from OpenAI into Meta’s superintelligence org (summer 2025 moves summarized in new reporting published Dec 26). (businessinsider.com)
- Named researchers (per published running list) and what they worked on previously:
- Jason Wei (worked on OpenAI o1 and “deep research” models) → joined Meta (July 2025). (businessinsider.com)
- Zhiqing Sun → joined Meta (July 2025). (businessinsider.com)
- Hyung Won Chung → joined Meta (July 2025). (businessinsider.com)
- Shengjia Zhao (co-created ChatGPT and GPT-4) → became chief scientist of Meta’s superintelligence lab (July 2025). (businessinsider.com)
- Jiahui Yu (led OpenAI “Perception” work enabling multimodal “senses”) → joined Meta (late June 2025). (businessinsider.com)
- Hongyu Ren (core contributor to GPT-4o) → joined Meta (summer 2025). (businessinsider.com)
- Shuchao Bi (multimodal + RL; focus on RL/post-training/agents at Meta) → joined Meta (June 2025). (businessinsider.com)
- OpenAI → created a new senior safety role (“Head of Preparedness”) and started recruiting
- Decision: OpenAI posted for a Head of Preparedness role inside its Safety Systems team (San Francisco).
- What the role is meant to do (per job description):
- Lead strategy/execution of OpenAI’s Preparedness framework and run a scalable “safety pipeline” combining capability evaluations, threat models, and mitigations across frontier-risk domains (e.g., cyber/bio). (openai.com)
- Compensation signal: $555K + equity (listed). (openai.com)
- Why it’s notable: This is a concrete org-level decision to add dedicated leadership capacity for frontier capability risk tracking, at a time of heightened public scrutiny of model harms. (openai.com)
Projects paused/stopped (frontier model orgs)
- OpenAI → “code red” internal reprioritization (temporary pausing of initiatives)
- Decision (reported Dec 29): Sam Altman reportedly instituted a temporary “code red” in early December 2025, telling staff to pause other initiatives to prioritize improving ChatGPT. (archive.ph)
- A specific initiative reportedly paused: Work on integrating ads into ChatGPT was reportedly paused as part of the “code red” (with insiders saying ads may still be planned for 2026). (archive.ph)
- What I did not find in this 4-day window: verified reporting that a major lab fully canceled a frontier-model training program (as opposed to pausing / delaying / reprioritizing).
Adjacent (but highly relevant): decisions impacting frontier-model economics (inference bottleneck)
- Nvidia ↔ Groq → licensing + apparent “acqui-hire” to accelerate inference
- Decision (reported Dec 29): Nvidia and Groq entered a non-exclusive inference technology licensing agreement aimed at making inference faster/cheaper; Groq founder Jonathan Ross and other employees (incl. Groq president Sunny Madra) are joining Nvidia, while Groq continues independently. (axios.com)
- Why it matters for frontier model orgs: This is a direct move at the inference bottleneck—the cost/performance constraint that increasingly determines whether frontier models can be deployed profitably at scale. (axios.com)
Ground-truth source links (URLs)
OpenAI — Head of Preparedness job posting: https://openai.com/careers/head-of-preparedness-san-francisco/
The Economist — “OpenAI faces a make-or-break year in 2026” (Dec 29, 2025): https://www.economist.com/business/2025/12/29/openai-faces-a-make-or-break-year-in-2026
Reuters — Meta to acquire Manus (Dec 30, 2025): https://www.reuters.com/world/china/meta-acquire-chinese-startup-manus-boost-advanced-ai-features-2025-12-29/
Manus — “Manus Joins Meta for Next Era of Innovation” (Dec 29, 2025): https://manus.im/blog/manus-joins-meta-for-next-era-of-innovation
The Straits Times — Meta to acquire Singapore-based start-up Manus (Dec 30, 2025): https://www.straitstimes.com/business/companies-markets/meta-to-acquire-singapore-based-start-up-manus-to-boost-advanced-ai-features//
Business Insider — departures from OpenAI incl. researchers lost to Meta (Dec 26, 2025): https://www.businessinsider.com/executives-board-members-and-researchers-who-left-openai-in-2025-2025-12
Axios — Nvidia/Groq inference licensing + hires (Dec 29, 2025): https://www.axios.com/2025/12/29/nvidia-groq-inference-chips
Published on December 30, 2025