Zhipu's GLM-5.4/5.5 roadmap leaks: 1T+ params and a full RSI self-training loop
airesearch12 · x · 2026-10-11
A leak attributed to Zhipu founder Tang Jie outlines the GLM-5.4/5.5 roadmap: scaling from 740B to over 1 trillion parameters, with the goal of a Fully Self-Training RSI loop where the model participates in its own iteration from pretraining through post-training.
- First major Chinese lab to explicitly join the recursive self-improvement race, shifting competition from raw scale to self-training
- Commenter's take: the trillion params are just infrastructure; the hard part is reliable self-evaluation of training quality, and real-world results remain to be seen
More from Models
- Anonymous unreleased AI model builds impressive pure-code three.js in hours — karminski3 · 2026-10-11
- Researcher Breaks Qwen 2.5 via Endless Gaslighting, Forced Off arXiv by Endorsement Rule — IndraVahan · 2026-10-11
- Outside CVP/Daybreak, the world's best cybersecurity model is Chinese GLM 5.3, not Claude — zephyr_z9 · 2026-10-11
- Was Claude's gibberish fixed by capping KL divergence in RL? One theory — burny_tech · 2026-10-11
- 20-year engineer benchmarks Gemma4-31B vs Qwen3.8-27B locally; GPT-6.1-Sol is still another tier — therealjerseytom · 2026-10-11
- Rumor: Google employees claim internal model Carbon beats unreleased Gemini — bindureddy · 2026-10-11