What Proximal does

Proximal AI, Inc. is a San Francisco-based research lab and data company, founded in February 2026, that builds AI training data and execution environments for software engineering agents and autonomous coding systems. The company operates two core products: the FrontierSWE benchmark, a public leaderboard evaluating agents on 17 ultra-long-horizon tasks requiring up to 20 hours per run across implementation, performance, and research categories; and the proprietary Proximal Platform, an internal data engine that synthesizes entire codebases, orchestrates multi-agent systems running 30+ days in parallel, and supports RL training infrastructure (GRPO, SDPO) against trillion-parameter open-source LLMs. Supporting components include a multi-node snapshotting system, universal rollout abstractions spanning serverless and full-parameter fine-tuning backends, agentic QA, and large-scale ingestion of 100M+ public PRs annually from GitHub, GitLab, Bitbucket, and Stack Overflow.

The company monetizes through bespoke B2B professional services delivered under a Master Development Services Agreement with project-specific Order Forms, priced per engagement with monthly invoicing and 30-day payment terms. Its primary target segment is AI/ML companies and frontier research labs building coding agents, with a secondary segment of enterprise customers seeking customized AI development services. Go-to-market combines direct enterprise sales for service contracts with a research-first open publication strategy — blog posts, the FrontierSWE leaderboard, and open-source repositories — used to attract talent, signal credibility, and pull in future enterprise buyers.

The team, drawn from Cursor, Prime Intellect, and Jane Street, operates from San Francisco and Bangalore, and is backed by top-tier venture funds and angels from OpenAI, Anthropic, xAI, Meta Superintelligence, Google DeepMind, and Thinking Machines. Proximal's competitive posture centers on the belief that data complex enough to train frontier models is bottlenecked by engineering ideas and software infrastructure rather than by domain-expert labeling, and on a shared data-engine architecture that compounds learnings across domains.

Name

Proximal

Legal name

Proximal AI, Inc.

Website

https://proximal.ai

Company type

Private

Founded year

2026

Operating status

Operating

Headcount range

11–50 employees

Short description

Proximal is a San Francisco-based AI research lab and data company that builds synthetic training data, agent execution environments, and RL infrastructure for frontier coding agents, and operates the public FrontierSWE benchmark for ultra-long-horizon software engineering tasks.

Ownership category

akta.pro rank

Product category

AI Training Data Services

SIC

Services-Computer Programming, Data Processing, Etc. (7370), Services-Prepackaged Software (7372)

akta.pro primary industry

AutoML & Low-Code ML Platform Operations (HDAAABAL)

Keywords

• AI training data
• Synthetic data generation
• Coding agent evaluation
• Autonomous coding systems
• Machine learning data services

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices3 records

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Technology or R&D, Infrastructure, Operations, Marketing or Sales

Revenue model

1. Custom Development Services: B2B professional services delivered via Order Forms under MSA. Services include custom Deliverables defined per customer engagement with acceptance testing and milestone-based delivery. Fees specified in Order Forms with monthly invoicing and 30-day payment terms.

Pricing tiers

Go-to-market motion2 records

Distribution channels1 record

Marketing channels4 records

Core offering

Proximal is a research lab and data company that builds AI training data, synthetic code generation pipelines, and evaluation environments to improve software engineering agents and autonomous coding systems. The company delivers custom development services to enterprise customers under Master Development Services Agreements, while also operating the FrontierSWE ultra-long-horizon coding benchmark to evaluate frontier coding agents.

Product overview

Proximal is a research lab and data company focused on advancing coding intelligence through training data innovation. The company operates two main offerings: the FrontierSWE benchmark platform for evaluating coding agents at the limits of human abilities, and the proprietary Proximal Platform for internal data generation and service delivery. The core focus is on creating high-quality training data for autonomous coding agents that can run for extended periods and solve complex technical problems.

Differentiator

Problem solved

Functional benefit

Products and services

• FrontierSWE Benchmark An ultra-long-horizon coding benchmark that evaluates AI coding agents on extremely difficult technical challenges requiring novel ideas and extensive planning. Includes 17 tasks across implementation, performance, and research categories, with agents given 20 hours per task and graded on a continuous scale.
• Custom Development Services

Quantifiable outcome

• Claude Fable 5 achieves 89% dominance (win rate vs random opponent) on FrontierSWE benchmark
• +2 more outcomes

Segments2 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability9 records

Feature8 records

Partnerships

Three partnerships are on record, tiered core.

• ModularcoreStrategic or Co-development Partner · 16 April 2026Academic and industry partner for FrontierSWE benchmark. Provided real-world problems from performance engineering domain, specifically contributing the 'Wan 2.1 on MAX/Mojo' task involving optimizing ML training with Mojo/MAX Graph implementation.
• Prime IntellectcoreStrategic or Co-development Partner · 16 April 2026Collaborative partner in FrontierSWE benchmark development, contributing expertise in RL and agent training. Team members have background from Prime Intellect where they built coding agents and RL infrastructure.
• Thoughtful LabcoreStrategic or Co-development Partner · 16 April 2026Industry partner contributing to FrontierSWE benchmark curation, providing problems from domains outside of Proximal's direct expertise to ensure benchmark diversity and reflects real engineering challenges.

Scale indicators5 records

Recent moves6 records

Expansion highlights6 records

Others

• Factory AI: Factory AI builds autonomous coding agents ('Droids') and is a probable Proximal customer. Comparable to Proximal in operating in the coding-agent space, though from the consuming/model side rather than the training-data side.
• Cognition (Devin): Cognition builds the Devin autonomous coding agent and is therefore a likely Proximal customer as well as a downstream beneficiary of better coding-agent training data. Adjacent rather than direct peer, but operates in the same coding-agent ecosystem.

Direct peers

• Gretel: Gretel provides a synthetic data platform for AI/ML training with APIs for generating tabular, text and time-series data. Comparable as a synthetic-data infrastructure peer, though less focused on long-horizon agent training specifically.
• Labelbox: Labelbox is a data-centric AI training platform combining labeling, evaluation and model diagnostics for enterprise and frontier customers. Comparable in providing the data infrastructure layer that AI labs rely on, though with more emphasis on human-in-the-loop labeling.
• Mostly AI: Mostly AI builds synthetic data generation technology for AI training, focused on privacy-preserving and statistically faithful data. Comparable to Proximal's synthetic-data-engine approach, though their domain is structured/tabular data rather than code.
• Tonic.ai: Tonic.ai provides synthetic data for software engineering and ML training, with strong roots in de-identification and realistic data generation. Comparable synthetic-data-generation capability though Tonic focuses more on test/de-identification than RL training data.
• Snorkel AI: Snorkel AI is a data-centric AI platform that uses programmatic labeling and synthetic data generation to build training datasets. Comparable to Proximal in the data-engineering approach to enabling AI model training, though focused more on enterprise ML than coding agents specifically.
• Surge AI: Surge AI provides high-quality training data and evaluation services specifically for frontier AI models, with strong emphasis on evaluation rubrics and benchmark-style work. Closest direct competitor in targeting the same buyer (frontier labs) with comparable data-quality rigor.

Emerging players

• Inworld AI: Inworld AI builds AI agents and uses synthetic data/simulation environments to train them, overlapping with Proximal's agent-training-infrastructure thesis. Less focused on code-generation agents specifically but directly comparable in the synthetic-data-for-agents framing.

Broad incumbents

• Scale AI: Scale AI is the largest data-labeling and synthetic-data provider for frontier AI labs. Most directly comparable to Proximal in terms of serving AI/ML customers with high-quality training data, but operates at vastly greater scale and breadth across modalities.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat3 records

Key risks6 records

Key highlights6 records

Customer concentration

• GitHubgithub.com/proximal-ai

Revenue estimate

Valuation estimate

Number of profiles

Profiles5 records

Funding overview

Funding rounds1 record

Investors1 record

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Proximal