Snorkel AI Highlights First Wave of Open Benchmarks Grants Projects
Snorkel AI has highlighted the first group of projects supported through its Open Benchmarks Grants program. The initiative aims to advance open-source datasets, benchmarks, and evaluation research for AI systems.
San Francisco – Snorkel AI has announced the initial projects receiving support through its Open Benchmarks Grants program, a $3 million commitment to foster open-source datasets, benchmarks, and evaluation research in artificial intelligence. The program launched in February 2026.
The initiative addresses the challenge of AI systems advancing faster than the field's ability to rigorously measure their performance on realistic tasks. Supported teams receive funding, expert data development assistance, and research and engineering collaboration.
Projects selected for this first wave include Frontier-Bench, designed as a more challenging and domain-diverse successor to previous evaluation tools, and Agents' Last Exam, which evaluates AI agents on long-horizon professional workflows across more than 55 industries.
Other supported projects aim to assess computer-use agents (OSWorld 2.0), measure genuine improvement in sequential tasks (Continual Learning Bench), and evaluate code quality degradation (SlopCode Bench).
Beyond the grants, Snorkel AI also led the development of Senior SWE-Bench, a benchmark for evaluating coding agents on senior-level engineering tasks. The Open Benchmarks Grants program also benefits from partnerships with organizations including Hugging Face and PyTorch, with applications being reviewed on a rolling basis.