jietang
15 Intel
Jie Tang is a Chair Professor of Computer Science at Tsinghua University and a co-founder of Z.ai (Zhipu AI), where he has led development of the GLM family of large language models. A Fellow of the ACM, AAAI, and IEEE, he built the academic search engine AMiner and has published more than 400 papers on data mining, social networks, and artificial general intelligence.
Is this you? Sign in with X to claim this profile.
Intel
Zhipu says GLM-5.3 Flash scored 57 on the Artificial Analysis index at roughly 1/100th of frontier model pricing while running on domestically made chips, and became OpenRouter's top model by weekly token share.
GLM-5.3 is coming. focusing on coding. https://t.co/g10V8eJIwq
Zhipu researchers describe Single-rollout Asynchronous Optimization (SAO), a reinforcement-learning method that stays stable for a thousand training steps and outperforms GRPO and its variants on SWE-Bench Verified, BeyondAIME, and IMOAnswerBench.
Zhipu's Jie Tang argues long-horizon agentic tasks are the next breakthrough. He says memory and continual learning are being approximated by long context and faster releases, self-judging is the tipping point, and self-training is the endgame. Several claims are speculative.
GLM-5.2 debuts at the top of open-source rankings on the Artificial Analysis Intelligence Index (score 51, matching Claude Opus 4.8), ranks #2 on Code Arena, #1 on DesignArena, and #3 globally on the FrontierSWE engineering benchmark.
Zhipu is rolling out GLM-5.1-highspeed, a pricier variant running at 400 tokens per second aimed at agentic workloads that need much faster generation than standard inference.
Inference providers are racing to serve GLM-5.2 faster, with reported throughput climbing from 280 to 318 tokens per second and Databricks later claiming 392, showing how much headroom remains in open-weight model serving.
Zhipu releases GLM-5.2 under an MIT license with a 1M-token context trained for coding-agent work. IndexShare attention cuts per-token FLOPs 2.9x at 1M, and it ranks highest among open models on FrontierSWE, PostTrainBench and SWE-Marathon.
1M users already https://t.co/lLKMytzbSk
Artificial Analysis Index = 60 https://t.co/ts3QfDtRmv
Zhipu announces GLM-5.2, its most capable open-weight model, with a 1M-token context window aimed at long-horizon agent tasks. It reaches GLM Coding Plan users first, with API access promised the following week. The post offers no benchmark numbers, so the capability claims are unverified.
A history of compute-optimal scaling from Kaplan's parameter-heavy exponents through Chinchilla's 20-tokens-per-parameter rule to today's deliberate over-training, driven by inference cost now dominating a model's lifetime spend rather than training cost alone.
ZCode is offering GLM-5.3-Flash free for 10 hours a day from September 3 to 18, prioritizing Coding Plan subscribers, with new users receiving a one-time 100M free token allotment during the window.
Zhipu describes how a GLM-5.3 Infra Agent helped bring GLM-5.3-Flash to production on new accelerators in two weeks with 3.2x throughput. The key lesson: layered, verifiable feedback beats slow end-to-end benchmarks. Humans still set goals and reviewed risky changes.
ZCode 3.0 launches optimized for GLM-5.2, adding stronger agentic task execution, better long-context coding, and a new Goal feature for managing multi-step development objectives end to end.