Skip to main content
AI Market Watch
Loading...

Proximal

Category: AI Infrastructure

Research lab for AI training data that builds long-horizon RL environments, evaluations and post-training data for frontier coding agents. The company is led by Calvin Chen. Based in San Francisco, United States. Team size: 11-50. Total funding raised: $15M. Latest round: Seed. Key investors include General Catalyst, Scribble Ventures, SV Angel, Chemistry, Go Global Ventures (Diede van Lamoen), Liam Fedus, Kevin Weil, Erik Bernhardsson.

Headquarters
San Francisco, United States
Team size
11-50
Total funding
$15M

Value proposition

Proximal treats training data as a research and engineering problem rather than a contractor or labeling problem. It builds software-driven data engines, RL environments grounded in real codebases, and verifiers. These produce complex, hard-to-solve coding tasks that measurably improve frontier models and agents.

Products and solutions

Long-horizon RL environments built from real codebases, coding post-training data and evaluations, systems that turn raw agent traces and human workflow artifacts into evals and targeted training data, FrontierSWE, an open-source ultra-long-horizon coding benchmark (v2 has 34 tasks, 13 models and 20-hour budgets), research on fuzzy verifiers and reward-hacking detection.

Unique value

A research-and-engineering-first data company. It automates data creation through synthetic generation, agentic QA and large-scale agent infrastructure instead of scaling human contractors, and publishes open benchmarks.

Target customer

Frontier AI labs, neolabs, top AI startups building coding agents, and enterprises adapting general models to niche use cases.

Industries served

AI/ML model development (frontier labs), software engineering. Stated expansion targets are drug discovery, chip design, energy and legacy critical-infrastructure software.

Technology advantage

Infrastructure runs very large numbers of parallel and long-duration (12+ hour) agent runs. Other advantages are synthetic codebase generation, PR-stream processing, fuzzy verifiers and reward-hacking stress tests. The team comes from Cursor, Google DeepMind, Meta Superintelligence, Prime Intellect, Citadel and Jane Street, and more than half are former founders.

How they differentiate

Incumbents such as Scale, Surge and Mercor scale through human contractors and QA pipelines. Proximal instead uses researchers and engineers to design task generation and verification at scale, focusing on complex long-horizon coding tasks and quality-aware (fuzzy) verification.

Main competitors

Mechanize, Mercor, Surge AI (also Scale AI, AfterQuery, Datacurve)

Key partnerships

General Catalyst (lead investor, Quentin Clark, Katie Keller). Works with frontier labs and enterprises, these are not publicly named.

Major milestones

Early 2026: company introduced by founders, Scribble Ventures-led early round (amount undisclosed). June 2026: 'Announcing Proximal' blog. April 2026: FrontierSWE benchmark released, followed later by FrontierSWE v2. 2026-09-29: emerged from stealth with a $15M seed led by General Catalyst at a $300M valuation, announced more than $200M annualized revenue, and began expanding beyond coding.

Growth metrics

Company claims more than $200M in annualized revenue about 10 months after starting, measured as quarter-to-date revenue ×4. Seed valuation was $300M. LinkedIn headcount band is 11-50.

Market positioning

Fast-growing specialist vendor of coding RL environments and data for frontier labs. It claims more than $200M in annualized revenue within about 10 months, measured as quarter-to-date revenue ×4, and is now expanding beyond software engineering.

Geographic focus

Global (US-headquartered; Bangalore presence)

Patents and IP

No patents found. The FrontierSWE benchmark is open source (GitHub Proximal-Labs/frontier-swe).

About Calvin Chen

Co-founder of Proximal; Y Combinator W23 founder (college dropout); previously bootstrapped and sold a SaaS order-management tool for dropshipping sellers while in high school; trained diffusion models for virtual try-on and worked on web agents before Proximal.

Latest news about Proximal

More AI Infrastructure companies

Official website: