Observed arrival · 2026-10-27
DataMining’s hard tasks for frontier models
DataMining presents expert-built task datasets for training and evaluating LLMs and agents, spanning coding, complex reasoning, multimodal work, and other fields.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- LLM and agent research teams
- Worth noticing
- Its Deep SWE-style comparison reports 15-percentage-point lower accuracy and 20-round longer agent trajectories than comparable provider data.
Field notes
The homepage groups its work across coding, complex reasoning, multimodal tasks, and agents, with legal, finance, healthcare, and security also listed as capability areas. It names benchmark references including ALE-Bench, HLE, Deep SWE, and APEX Agents, while saying its datasets are designed around benchmark difficulty rather than copied from benchmarks. Its performance figures are labeled as internal comparisons.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue