TokenRhythm Open-Sources NeoHorse-1: Feeding Router Trajectories to Small Models for a Verified RSI Loop
机器之心 · wechat · 2026-09-09
TokenRhythm has open-sourced NeoHorse-1 (4B and 9B, built on Qwen3.5). The core idea: let the routing system (OpenSquilla), originally built to dispatch work across models, also record every execution trajectory, then use those trajectories to post-train a small model so it learns to complete what a whole model pool once did. The team calls this a reproducible single-round validation of an RSI (Recursive Self-Improvement) loop, with Tsinghua, PKU and Infinigence involved.
Training proceeds in four steps: preserve full reasoning-action-feedback trajectories (100K–1M anonymized traces); filter quality via deterministic rules plus six-dimension semantic evaluation; perform curriculum learning using the router's predicted capability tiers (C0–C3); and apply router-guided On-Policy Distillation, with the next data mix steered by a "capability gap map." On evaluation, the 4B model improved from 58.94 to 64.87 (+5.93) and the 9B from 65.60 to 69.04 (+3.44), with the 4B nearing the Qwen3.5-9B base; gains concentrate on tasks with clear workflows and verifiable feedback, while long-horizon tasks remain hard.
The author cautions this is not full RSI—only one evaluate-select-update round is validated, and cross-generation gains remain unproven—but its significance is framing RSI as a reproducible engineering problem: a continuously running Harness that leaves structured records becomes the cheapest and hardest-to-copy training data for next-generation agent models.
More from Models
- Tencent Hunyuan details Gander, an end-to-end full-duplex omni interaction agent — Tencent-Hunyuan · 2026-09-09
- Moonshot open-sources MoBA, a block attention mechanism it says is 16x faster for long context — gekobraa · 2026-09-09
- From Failing 2+2 to PhD-Level Math: A Timeline of AI's Three-Year Leap — haider1 · 2026-09-09
- Muse Spark 1.3 reportedly crushes tax agent bench at $0.32/task, 3x cheaper — zainhas · 2026-09-09
- Does Step4 even exist? One "distillation" slide sparks debate on Chinese labs' training routes — zephyr_z9 · 2026-09-09
- Free API routes for GLM, DeepSeek Flash and more, plus a map of hidden AI free tiers — Roger_M_Taylor · 2026-09-09