Flash Update: Reflection AI Raises $2B to Build US Open Frontier AI Lab, Taking on DeepSeek

Flash Update: Reflection AI Raises $2B to Build US Open Frontier AI Lab, Taking on DeepSeek

TechCrunch – October 10, 2025 – Beijing

image

---

image

Image credit: Reflection AI

---

$2B Funding at $8B Valuation

Reflection AI — a startup founded in 2024 by two former Google DeepMind researchers — has closed a $2 billion funding round, bringing its valuation to $8 billion. That’s a 15× jump from its $545 million valuation just seven months ago.

Investors include:

  • NVIDIA
  • Disruptive
  • DST
  • 1789
  • B Capital
  • Lightspeed Venture Partners
  • Singapore’s Government Investment Corporation
  • Eric Yuan
  • Eric Schmidt
  • Citibank
  • Sequoia Capital
  • CRV
  • ...and others.

---

From Autonomous Coding to Open-Frontier AI

Originally, Reflection AI was focused on building an autonomous code-generation Agent. It now positions itself as:

  • An open-source alternative to closed frontier labs like OpenAI and Anthropic
  • A Western counterpart to Chinese AI leaders like DeepSeek

Founders:

  • Misha Laskin – led reward model development for DeepMind’s Gemini project
  • Ioannis Antonoglou – co-creator of AlphaGo, which beat the Go world champion in 2016

Their mission: prove top AI talent can create frontier-level models outside big-tech ecosystems.

---

Strategic Moves and Team

  • ~60 members: mostly AI researchers & engineers specializing in infrastructure, data training, and algorithms
  • Powerful compute clusters acquired
  • Planned release: frontier language model in 2026 trained on trillions of tokens
  • Claimed milestone: LLM + reinforcement learning platform for massive Mixture-of-Experts (MoE) models at frontier scale

---

Architectural Breakthrough: MoE

MoE (Mixture-of-Experts) architecture enables advanced LLMs — previously limited to large closed labs.

Chinese breakthroughs:

  • DeepSeek – first to openly train such models at frontier scale
  • Tongyi Qianwen and Kimi – following successes

> “If we don’t act, others will set the global intelligence standard,” — Misha Laskin

> “It won’t be the U.S.”

---

Geopolitical Angle

Laskin notes:

  • Avoiding Chinese models is common due to legal risks
  • U.S. must face the challenge or risk falling behind

---

Industry Response

David Sacks (White House AI & crypto affairs lead):

> “Great to see more U.S. open-source AI. Many global markets value the cost savings, customizability, and controllability of open source.”

Clement Delangue (CEO, Hugging Face):

> “Major boost for U.S. open-source AI.”

> “Next step: show rapid sharing for models & datasets.”

---

AiToEarn’s Role in Open-Source Ecosystems

In this rapidly evolving space, platforms like AiToEarn官网 help creators, researchers, and AI enthusiasts:

  • Generate AI-driven multi-platform content
  • Publish simultaneously on Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X
  • Access integrated analytics & AI model rankings (AI模型排名)

---

Reflection AI’s “Openness” Strategy

Similar to Meta’s Llama and Mistral:

  • Model weights will be public
  • Most datasets & training processes will remain proprietary

> “Model weights are the most useful — anyone can fine-tune them,” — Laskin

Business Model:

  • Free use for researchers
  • Revenue from:
  • Enterprises building products
  • Governments developing sovereign AI

---

Roadmap

  • First model: text-focused, multimodal planned
  • Funding allocated to computing resources
  • Launch expected: early 2026

---

References

TechCrunch article

---

Community & Opportunities

image

🚀 Recruiting next batch of interns

image

🚀 Seeking creative post-2000 entrepreneurs

image

image

---

About Z Potentials

image

---

AiToEarn in the Bigger Picture

Platforms like AiToEarn官网 facilitate:

  • Cross-platform AI-powered publishing
  • Content monetization
  • AI model performance tracking (AI模型排名)

With the rise of sovereign AI projects and creator-driven innovation, AiToEarn enables both professionals and hobbyists to maximize value from open AI resources.

---

-----------END-----------

Read the original

---

Would you like me to also create a summary version of this rewritten piece so it’s suitable for rapid publication in a newsletter? That could make the key funding, strategy, and geopolitical points instantly clear.

Read more

Translate the following blog post title into English, concise and natural. Return plain text only without quotes.

ChatGPT Atlas 发布,AI 浏览器大乱斗...

Translate the following blog post title into English, concise and natural. Return plain text only without quotes. ChatGPT Atlas 发布,AI 浏览器大乱斗...

# AI Browsers: When LLM Companies Step In 原创 lencx · 2025-10-22 07:00 · 上海 --- ## Overview Large Language Model (LLM) companies are making moves into the **AI browser** space. From new entrants like **Dia**[1], **Comet**[2], and **ChatGPT Atlas**[3], to established browsers like **Chrome** and **Edge** (which now feature

By Honghao Wang