AfterQuery reaches $3.2B valuation five months after Series A

×

AfterQuery has reportedly reached a $3.2 billion valuation, five months after announcing a $30 million Series A at a $300 million valuation. The San Francisco startup builds training data and reinforcement-learning environments for AI systems. It has not disclosed how much capital was involved in the newer round.

The valuation implies a more than 10-fold increase from the company’s April 2026 financing. Y Combinator partner Gustaf Alströmer described the pace as the fastest launch-to-unicorn progression in the accelerator’s history. AfterQuery’s founders are 22 and 23 years old and took part in Y Combinator’s Winter 2025 cohort.

The company’s own Series A announcement and technical description provides more detail about its work. AfterQuery says it had exceeded a $100 million annualized revenue run rate by September 1, 2026, only a few months after closing that Series A. That is a run-rate figure, not reported revenue for a completed fiscal period, and the company did not provide a profit figure.

Training professional behavior, not just factual answers

AfterQuery is targeting a problem beyond conventional question-and-answer evaluation. Its datasets and reinforcement-learning environments are designed to capture how specialists make decisions and complete workflows, including judgment calls and edge cases that are difficult to extract from public text.

The company describes the objective as “encoding the patterns, decisions, and reasoning of the world’s best practitioners.” In practical terms, that means collecting examples and evaluations from people who work in fields such as engineering, medicine, law and finance, then using software to structure that material for model training and assessment.

“The knowledge that matters most is the hardest to capture.”

AfterQuery says its systems are used to encode, evaluate and scale expertise across domains including financial reasoning, software engineering and enterprise workflows. It also says leading AI labs use its datasets and environments. Publicly named customers include Nvidia, Legora and Motif Technologies, a Korean AI lab.

A dataset that checks whether a model produced the right answer is different from an environment that measures whether it followed a sound professional process. The latter can include intermediate decisions, tool use, exception handling and end-to-end task completion, although AfterQuery’s announcement does not publish the schemas, task counts, annotation costs or benchmark results behind those environments.

The company argues that this data layer is becoming more valuable as foundation-model architectures and research techniques spread. Its stated thesis is that high-quality expert data and evaluation environments can expose failure modes that are invisible on broad academic tests or synthetic tasks. AfterQuery says the relevant workloads include structuring billion-dollar mergers and acquisitions, designing personalized treatment plans and synthesizing large bodies of case law.

Those examples are ambitions, not evidence that the company’s systems can safely perform those jobs. No independent benchmark, model improvement figure or customer spending breakdown accompanied the valuation report or the company’s announcement. The public material also does not say which models were trained, how much of the work uses reinforcement learning rather than supervised data, or how expert judgments are reconciled when professionals disagree.

A software-controlled expert network

AfterQuery says its operational model is “research-driven” and “software-first.” Rather than outsourcing data collection as a generic labeling operation, it says it builds custom tools and workflows for each project and manages creation internally. The stated goal is tighter quality control and a better workflow for verified contributors.

The network now includes nearly 100,000 verified practicing professionals across engineering, medicine, law, finance and other fields. That number describes the size of the available network, not the number of contributors to any one dataset or the volume of completed training examples.

This approach puts the company closer to the expert-data businesses that have emerged around model development than to a conventional software-as-a-service vendor. The source material places AfterQuery in the same broad category as Mercor and Scale, but says its focus is not simply verifying factual answers. It is trying to represent specialists' work patterns well enough for models and agents to reproduce parts of those workflows.

The model also creates a direct cost and quality-control challenge. Professional contributors are more expensive and harder to coordinate than general-purpose annotators, while domain-specific evaluations can become obsolete as tools and workflows change. AfterQuery acknowledges that each professional domain has its own failure patterns and that completing workflows end to end is not a problem solved once.

Its April financing was led by Altos Ventures, with participation from The Raine Group and existing investors Y Combinator, BoxGroup and Latitude Capital. The company also says it is backed by angel investors from Google DeepMind, OpenAI, Anthropic, Meta Superintelligence Labs and Microsoft AI. It is hiring in operations, engineering and research as it expands the expert network and domain coverage.

There is one timeline discrepancy in the public account. AfterQuery’s announcement says the company was 14 months old when describing its progress, while the financing coverage places its founders in the Winter 2025 Y Combinator cohort about 18 months before September 1, 2026. Those figures may refer to different milestones, but the company has not clarified the distinction. The interval between the $300 million Series A valuation in April 2026 and the reported $3.2 billion valuation on September 1 is unusually short; the size and investors in the latest round remain undisclosed.