Xiaohongshu's Vela 35B turns anonymized user behavior into supervision, gains +11.5 on VelaBench
AdinaYakup · x · 2026-10-16
Xiaohongshu (RedNote) is doing serious research, not just shipping models. Vela uses anonymized user behavior as supervision: behavior histories from 100K users are converted into tasks across 91 templates, filtered down to 75K training examples. Fine-tuning Qwen3.6-35B-A3B improves over the base on all 10 benchmarks — +11.5 pts on VelaBench, +6.2 avg on four user-related benchmarks, and +4.0 avg on five general benchmarks. The work targets long-term context learning, evolving user intent understanding, and behavior-grounded decision-making for complex decisions, long-horizon planning and autonomous agent execution. Behavior records were anonymized, platform-authorized, and passed privacy/ethics review.
More from Models
- Open source is so back: dozens of open models dropped within 48 hours — beffjezos · 2026-10-16
- Gemini users hit usage caps on 3.1 Pro and 4 Argon, resets not until Oct 16 — R3d-Gr33n-Blu3 · 2026-10-16
- Gemini desktop app can't accept images — an image model that can't take images — Rebl · 2026-10-16
- Viral screenshot claims GPT-6 Astra proved a physics benchmark wrong, lifting its score to 98% — Confident_Salt_8108 · 2026-10-16
- Apollo Research: GLM-5.3 unprompted reasons about its actions becoming training data — _lewtun · 2026-10-16
- Opus 5.5 turns heads with surprisingly polished vertical video demo — FinanceYF5 · 2026-10-16