TokenRhythm Launches NeoHorse-1: 4B/9B Models Post-Trained on Agent Execution Traces
rohanpaul_ai · x · 2026-09-17
TokenRhythm released its first in-house model family, NeoHorse-1, in 4B and 9B variants, both starting from Qwen3.5 checkpoints.
Key ideas:
- Most agent systems throw away their best training data after every run: which model got routed where, which tools were called, what came back, where the agent failed, and how it recovered.
- NeoHorse-1 is post-trained not on Q&A but on structured execution trajectories generated inside TokenRhythm's OpenSquilla Harness — capturing when to search, when to use a tool, what failed, and how the system got back on track.
- The team frames this as an early engineering validation of RSI via two connected loops: Data-RSI + Model-RSI, under the thesis that models won't converge into one winner — agents need to organize models.
More from Models
- Open-source VoxLock tops Krisp voice isolation benchmark with 4.95% WER — Arbaz Siddiqui · 2026-11-02
- Anthropic's 85% inference margin sparks debate on open source's price-pressuring role — code_star · 2026-11-02
- User reports xAI appears to have quietly raised weekly usage limits, dropping his meter from 93% to 47% — XFreeze · 2026-11-02
- Meituan's LongCat-2.5 lands as 1.6T MoE agent model, with DeepSeek copycat speculation — teortaxesTex · 2026-11-02
- Asked what AI firms want you to stop owning, Claude said: your own compute — alex_verem · 2026-11-02
- Is Hermes Still Worth It? Codex's $200 Tier Now Overlaps Most of Its Value — Foreign_Lead_3582 · 2026-11-02