Realset AI and Flatkey Raised $10 Million
The companies secured Series A funding to expand data collection for large language models and embodied AI.
Updated on Sept. 30, 2026 in Artificial Intelligence
Live Poll
Do you believe increased investment in AI training data will improve future technology safety?
Realset AI and Flatkey have raised $10 million in Series A funding to scale their production of training data for frontier models and embodied agents. This capital will support the expansion of their workplace and studio environment capture networks.
Why it matters
The investment accelerates the development of specialized training datasets for complex AI environments, which are essential for improving the reasoning and physical interaction capabilities of next-generation models.
The companies secured $10 million in Series A funding to support their training data pipelines. The capital is earmarked for new benchmarks intended to measure AI policy performance.
The players
Realset AI
A data engineering firm specializing in the creation of synthetic and observational training datasets for frontier AI models and embodied robotic systems.
Flatkey
A collaborative partner firm focused on data acquisition infrastructure for large-scale machine learning training pipelines.
The details
Realset AI produces specialized training data designed to refine the behavior of Large Language Models (LLMs) — AI systems trained on vast text corpora — and embodied agents — autonomous systems designed to interact with physical environments. By expanding their capture network of studio and workplace environments, the firms aim to increase the diversity of real-world scenarios available for training. The process also includes growing a pool of human expert demonstrators to provide high-quality feedback loops for model alignment.
Timeline
September 30, 2026: The funding round was officially announced.
The Tech Race
The funding follows a broader industry trend toward the creation of standardized open AI safety benchmarks to measure model performance. This investment directly extends that focus by funding the infrastructure needed to generate standardized evaluation data.
The development will likely result in more capable AI models that demonstrate better performance in workplace and office-based tasks. Users can monitor the project by tracking the eventual release of the promised open benchmarks for AI policy performance.
The takeaway
As model training pivots toward physical and professional environment data, the ability to generate standardized benchmarks will be a critical differentiator for AI providers. Watch for the forthcoming release of their open policy benchmarks to assess how these datasets improve model reliability.
Further reading
For more on the data infrastructure supporting current models, visit our Artificial Intelligence section.
Live Poll
Do you believe increased investment in AI training data will improve future technology safety?
