Datacurve

Datacurve provides research and data collection infrastructure for training and evaluating AI models. It addresses the challenge of capturing expert judgment and real-world task complexity so that models can learn from high-quality, task-faithful data rather than simplified examples.

Its product line includes reinforcement learning environments for measuring agentic capabilities across domains such as software engineering, data science, cyber security, machine learning, and research; long-horizon tasks spanning hours or days; prebuilt OTS datasets structured for direct use in training stacks; benchmarks and evals designed to reflect domain-sensitive performance; agent trajectories recording full expert execution traces including tool calls and recoveries; and supervised fine-tuning demonstrations captured through bespoke tooling. The DeepSWE benchmark is also listed as a product.

Datacurve is aimed at teams building and evaluating AI models who need curated datasets, evaluation environments, and expert task traces. The pages describe products and research areas but do not specify packaging details such as free plans, trials, editions, or pricing models.