Proximal
AI training data and post-training infrastructure
CORE INFO
Proximal builds systems that help frontier AI labs and enterprises turn real-world data and agent activity into evaluations and targeted training data. Its work currently centers on improving coding agents and other long-horizon AI systems.
- $15 million seed round led by General Catalyst at a reported $300 million valuation, announced in September 2026.theinformation.comwsgr.com
- Annualized revenue exceeded $200 million according to a company-reported figure that has not been independently verified.proximal.ai
- The Information reported on Proximal's seed financing, $300 million valuation, founders, and early revenue context in September 2026.theinformation.com
- FrontierSWE benchmark evaluates coding agents on 34 difficult tasks across 13 frontier models, with up to 20 hours allowed per task.proximal.ai
WHY WE WOULD WORK AT PROXIMAL
Solve AI's Data Bottleneck
Build the feedback loop that turns real-world data and agent activity into evaluations, capability insights, and targeted training data for frontier models.
Work on Frontier Systems
Develop large-scale rollouts, reinforcement learning, synthetic codebases, reward-hacking detection, and long-horizon agent training across Kubernetes and multi-GPU environments.
Shape Coding Agents
Contribute to FrontierSWE, Proximal's 20-hour coding benchmark spanning compiler optimization, inference systems, computational science, and research-grade software engineering.
Join a High-Ownership Team
Work alongside a roughly 40-person team whose backgrounds include Cursor, Google DeepMind, Meta Superintelligence, Jane Street, and Citadel. More than half the team are former founders.
Grow at the Ground Floor
Join a company founded in 2025 as it expands from coding agents into drug design, chip design, healthcare infrastructure, power grids, and government systems.
Backed for Ambitious Work
Help turn a $15 million seed round led by General Catalyst into foundational AI infrastructure, with the round reportedly valuing Proximal at $300 million.
MARKET AND TRACTION
MARKET POSITION
GROWTH TACTICS
KEY METRICS
✦ KEY METRICRESEARCH COLLABORATORS
COMPETITIVE ADVANTAGE
PRODUCT AND TECH
FrontierSWE Benchmark
Proximal's long-horizon coding-agent benchmark evaluates difficult implementation, performance-engineering, and research tasks with up to 20-hour execution budgets. It measures partial progress, functional coverage, and performance rather than simple pass/fail results.
Evaluation Systems
Proximal converts real-world data and agent activity into evaluations, rubrics, verifiers, and capability-gap analyses. These systems help identify where models fail and assess the usefulness of individual training examples.
Synthetic Data Generation
Proximal develops synthetic codebases, agentic tasks, rubrics, and verification data for targeted model improvement. Its approach automates data creation for technically difficult software-engineering and coding-agent scenarios.
Post-Training Infrastructure
Proximal builds reinforcement-learning and post-training systems using large-scale rollouts, agentic judges, reward signals, and reward-hacking detection. Its infrastructure supports long-horizon agent training and multi-agent software-development simulations.
Agent Execution Infrastructure
Proximal operates distributed systems for parallel, long-running agent processes across Kubernetes clusters, multi-GPU environments, and cloud execution environments such as AWS. It also supports isolated environments and multi-node snapshotting of disk, process memory, and GPU state.