Ninja Subnet Releases Public Dataset
const_reborn · x · 2026-07-15
Ninja Subnet released a new public dataset containing: - 600 qualifying tasks - 1,675 filtered-out candidate tasks - A total of 2,275 task samples - 44,752 sets of qualifier and battle rollouts from 6 retired kings They stated that the dataset will continue to grow alongside the competition: whenever a new king emerges, the tasks and rollouts of the retired king will be automatically archived. As complete LLM traces accumulate, this dataset will become increasingly useful. The team also mentioned future plans to explore collaborations with model training subnets, using SFT, preference optimization, and reinforcement learning to post-train models, hoping to make Ninja the preferred agent harness for these models.
More from coding & agent
- AI makes software easier to build, but it also lowers the floor on quality — paw_lean · 2026-07-21
- Fable coding run costs $6.69 for 67 lines of code in a 4-minute job — bytebot · 2026-07-21
- Cursor writes better code, but ChatGPT can still control the computer — vista8 · 2026-07-21
- Agents can remember facts, but still forget how to do the job — No_Advertising2536 · 2026-07-21
- Agent skills for project downgrade and troubleshooting tested in CLAD on LS 5.22 — stspanho · 2026-07-21
- Open-source B-roll skill turns scripts into 5-second vertical clips with Codex and Gemini — yangyi · 2026-07-21