Datacurve

Datacurve builds the research and data collection infrastructure to teach AI mastery across complex knowledge work.

Datacurve builds the data engine and research infrastructure required to train frontier artificial intelligence models. The company specializes in creating high-quality, long-horizon datasets and reinforcement learning environments that capture expert human judgment and mastery. Its platform turns complex, specialized knowledge into standardized, structured inputs for foundational AI labs and large enterprises.

The company's core offerings include supervised fine-tuning (SFT) demonstrations, reinforcement learning environments for agentic capabilities, and off-the-shelf (OTS) datasets curated for signal. Datacurve also develops advanced benchmarks, such as DeepSWE, which evaluates coding agents on original, long-horizon engineering tasks across multiple programming languages.

Based in San Francisco, Datacurve operates as a research-driven organization focused on pushing the boundaries of what machines are capable of. Its infrastructure is designed to be self-healing and deterministic, providing the data acquisition and enrichment pipelines necessary for the next generation of AI models.

Industries: Artificial Intelligence, Data & Analytics

Categories: Data Labeling, AI Infrastructure

AntSpace put this profile together from public information. It may be incomplete; once Datacurve claims it, its admins can edit everything here.

Our culture

Datacurve describes its team as a small group of intense, driven, and relentlessly curious technologists. The work environment is characterized as fast, focused, and quietly ambitious, where team members have the autonomy and intention to ship work that matters at a world scale. Employees are expected to be high-agency generalists who can also develop intense specialties in data-intensive systems. The company values systems thinking, product intuition, and an obsession with reliability and craftsmanship. It encourages independent, iconoclastic thinking and a willingness to question consensus. Work involves long days on systems and solutions that compound, with a focus on solving the hardest problems stunting model improvement side-by-side with researchers.

Core values

  • High agency
  • Obsession with reliability
  • Product intuition
  • Quality craftsmanship
  • Autonomy

Specialties

  • Reinforcement learning environments
  • Data acquisition pipelines

Markets served

  • Enterprise Software

Key facts

Legal name
Datacurve

Latest from Datacurve

  • Added GPT-6 Astra results (Project update · September 3, 2026)
    Datacurve added performance results for the GPT-6 Astra model to its coding benchmark leaderboard, covering various reasoning effort levels from low to maximum.
  • DeepSWE v1.1: cleaner and more reproducible (Product launch · June 14, 2026)
    The company released an updated version of its long-horizon engineering benchmark, introducing isolated verification in fresh containers and structured test reports for better reproducibility.
  • Introducing DeepSWE (Product launch · May 26, 2026)
    Datacurve introduced DeepSWE, a new benchmark designed to evaluate frontier AI coding agents on complex, long-horizon software engineering tasks across five programming languages.
  • Introducing DeepSWE Benchmark Research (Company update · May 18, 2026)
    The company published research on Deep SWE, a benchmark featuring original tasks with reference solutions averaging 668 lines of code to probe the limits of frontier models.

All 4 updates

Services by Datacurve

All 1 services

AI models by Datacurve

All 4 AI models

Documents

1 document

All 1 documents

3 more products, services and AI on AntSpace. Sign up free to see the full catalog.

Leadership, clients, case studies, documents and every location on AntSpace. Sign up free to see them.

Related