SourceScore

Topic hub · 23 claims

AI organizations — labs, founders, and the talent map

The labs and companies that ship frontier AI/ML. Founding dates, parent organizations, and the lineage that shaped each lab's culture. Hand-verified from official corporate pages + press records.

Mapping the lab landscape

Frontier-AI work in 2026 is concentrated across roughly a dozen labs: OpenAI, Anthropic, Google DeepMind, Meta AI, Mistral, Microsoft Research, xAI, Stability AI, Cohere, AI21 Labs, EleutherAI, Together AI, plus the Chinese frontier (DeepSeek, Alibaba/Qwen, Zhipu/GLM). Each has its own model lineage, alignment philosophy, and funding model.

Founders + lineage matters

Anthropic's founders left OpenAI in 2021 over alignment-direction disagreements. DeepMind was acquired by Google in 2014 but maintained a distinct research culture until the 2023 merger with Google Brain. Mistral was founded by ex-DeepMind + ex-Meta researchers in 2023. Knowing where each lab's researchers came from helps predict what kind of models they'll ship.

Open-source vs closed

The labs split roughly into open-weight (Meta, Mistral, Stability AI, EleutherAI, Hugging Face, Together AI, Alibaba's Qwen, DeepSeek, Allen Institute's OLMo) and closed-API (OpenAI, Anthropic, Google's Gemini API, Cohere, xAI). Some are hybrid (Google releases Gemma weights but not Gemini's). The boundary moves: 2024 saw multiple closed labs release smaller open-weight variants under pressure from open competitors.

Defined terms (3)

Frontier lab
An AI lab that produces models at or near the current state-of-the-art for general-purpose capabilities. As of 2026: OpenAI, Anthropic, Google DeepMind, Meta AI, Mistral, plus a small number of Chinese labs.
Open-weight model
A model whose trained weights are publicly downloadable, regardless of whether training data + code are also released.
Lab lineage
The chain of researcher movements that shape a lab's culture and research direction. Often more predictive of model behavior than corporate stated priorities.

All claims in this topic (23)

Related

Concept pillars

Framework integrations