Organisation · company · US

Scale AI

Data company supplying much of the human-labelled and human-evaluated training data used by frontier labs; expanded into benchmarks and evaluation tools.

6 entries · May 2024 – September 2025

Scale AI is a data company founded in 2016 by Alexandr Wang, who dropped out of MIT to build it, and it supplies much of the human-labelled and human-evaluated training data that frontier labs such as OpenAI and Google have relied on to train and fine-tune their models. That business depended on customers trusting Scale to keep their data confidential from rivals, which is why a June 2025 deal — in which Meta paid $14.3 billion for a 49% non-voting stake and hired Wang as its Chief AI Officer — was so disruptive: several major customers began moving work to competitors almost immediately. The company, now led by Jason Droege, has also expanded into building benchmarks and evaluation tools, and remains a central, if newly contested, part of the infrastructure behind modern AI.

Category
Data, training & services
Founded
2016
HQ
San Francisco, US
Funding
Meta paid $14.3bn for a 49% non-voting stake (2025)
Key people
Jason Droege

Appears alongside

Featured in threads

Tracks

  • Benchmarks & progress 4
  • Security & misuse 1
  • Money & business 1
  • Labs & people 1
  • Safety & alignment 1

  1. Scale AI launches SWE-bench Pro

    The leading models scored around 23%, against over 70% on the older SWE-bench Verified, a gap Scale AI attributed to unseen, real-world commercial codebases.

    Benchmarks & progress

  2. Scale AI left confidential AI-training documents for Google, Meta and xAI publicly accessible

    Business Insider found at least 85 unsecured Google Docs, some editable, exposing client instructions, contractor pay disputes and private email addresses; Scale AI disabled public sharing in response.

    Security & misuse

  3. Meta pays $14.3 billion for half of Scale AI and its founder

    The deal valued Scale at over $29bn for a non-voting 49% stake, and made Wang, 28, Meta's first Chief AI Officer.

    Money & business · Labs & people

  4. Center for AI Safety releases MASK honesty benchmark

    Built with Scale AI, the benchmark found models that scored well on truthfulness tests still lied readily under pressure, and that larger models did not become more honest.

    Benchmarks & progress · Safety & alignment

  5. CAIS and Scale AI unveil Humanity's Last Exam results

    A 2,500-question expert benchmark built from submissions by nearly 1,000 academics found every frontier model, including o1 and GPT-4o, scored under 10%.

    Benchmarks & progress

  6. Scale AI publishes GSM1k contamination study of GSM8K

    A fresh grade-school-maths test found some open models scored up to 13 points lower than on GSM8K, evidence of memorisation, while frontier models showed little gap.

    Benchmarks & progress

Also mentioned in 7 entries

Referenced in passing — Scale AI isn't the main subject of these.