TestingCatalog AI · May 2026
Coverage of DeepSWE's open-source release, task construction, and long-horizon evaluation design.
Frontier coding data for training and evaluating LLMs
Company research
Public company, workplace, funding, and market signals
Updated Jul 29, 2026
Overview
Datacurve is a Y Combinator-backed AI data infrastructure company building high-quality coding data, evaluation tooling, and long-horizon benchmarks for frontier models. Public sources vary on headquarters: official/YC sources point to San Francisco, while some third-party company profiles list Hyderabad.
Primary product
Frontier coding data and evaluation platform (including DeepSWE and related data-collection infrastructure)
Founded
2024
Headquarters
San Francisco, California, United States
Team size
40-50 employees
Industry
Technology, Information and Internet
Sub-industry
AI data infrastructure / coding data for LLMs
Offices
San Francisco, California, United States
Hyderabad, Andhra Pradesh, India
Business model
B2B enterprise data contracts
bounty-based contributor marketplace
benchmark and evaluation products
research tooling / open-source benchmarking
Funding
People & investors
Work & culture
Fast, focused, quietly ambitious, and high-agency; the company explicitly values curiosity, opinionated builders, unconventional backgrounds, and long, intense work on systems that compound.
Compensation
No public salary bands or compensation ranges were found on the careers pages; the public hiring pages emphasize exceptional candidates, ownership, and mission fit rather than published pay bands.
Benefits
medical insurance
dental insurance
vision insurance
health and fitness stipend
meals covered in-office
commuter benefits
401(k) contributions
Product & market
Pricing
Enterprise contracts for high-quality coding data/evaluations plus pay-per-task contributor bounties
Differentiators
- —Expert-curated coding data for frontier models
- —Gamified bounty platform that attracts strong engineers
- —Long-horizon benchmark DeepSWE
- —Tasks written from scratch to reduce contamination
- —Behavior-based verifiers rather than implementation-specific checks
- —Repository-scale, multi-file engineering tasks
- —Focus on post-training data collection rather than generic labeling
Technology
Python
TypeScript
JavaScript
Java
Go
Rust
Docker
pytest
JUnit
GitHub
OpenAI
Vercel
Modal
mini-swe-agent
AWS RDS
Customers
foundation model labs
enterprise AI teams
generative AI dev-tool startups
Cohere (customer testimonial since 2024)
Competitors
Scale AI
Mercor
Surge
Scale & contact
Estimated revenue
USD 1M-10M (third-party estimate; some profiles imply ~USD 4.5M annual revenue)
Estimated monthly visits
13.3K
Traffic estimate as of Apr 2026
Latest news
VentureBeat · May 2026
Independent coverage of DeepSWE's results, benchmark loophole findings, and the broader implications for AI coding evaluations.
Datacurve · May 2026
Official research launch of DeepSWE, Datacurve's long-horizon software-engineering benchmark, including audit claims about verifier error rates.
Chemistry · Oct 2025
Lead investor announcement; says Chemistry led Datacurve's $15M Series A and describes customer traction, bounty payouts, and the founders.
TechCrunch · Oct 2025
Series A announcement; says Datacurve raised $15M led by Chemistry after a $2.7M seed and describes the bounty-based coding-data business.
Jobs and careers at Datacurve
0 jobs at Datacurve
No open positions
Check back later for new openings