ModelsSelected

GLM-5.3 上线 SambaCloud,推理速度超 200 tok/s

Why we picked it

智谱的 GLM-5.3 现在在 SambaCloud 上跑出 200+ tok/s,Terminal-Bench 3.0 拿了开源权重里的最高分 28.3。用它完成任务比几个闭源前沿模型还省 token。写脚手架的、跑 Claude Code 或 Codex CLI 的可以直接换上试试,成本能压不少。

智谱的 GLM-5.3 现已通过 SambaCloud 提供,推理速度超过 200 tok/s。该模型在 Terminal-Bench 3.0 拿到 28.3 分、DeepSWE v1.1 拿到 66.9 分,是这两个编码基准上的开源权重最高水平。

Body

成绩的一个细节是,GLM-5.3 完成任务所需的输出 token 数少于若干闭源前沿模型,意味着在同等基准下生成更少内容就能达标,实际使用中能降低延迟和 token 成本。

开发者可以直接把 Claude Code、Codex CLI 或 ZCode 指向 SambaCloud 的接口,把 model 参数设为 "GLM-5.3" 即可接入。开源权重模型在编码智能体基准上逼近闭源模型,对预算敏感的团队是更可行的选择。

Original · SambaNova

GLM-5.3 @Zai_org on SambaCloud: 200+ tok/s.

SOTA open-weight on Terminal-Bench 3.0 (28.3) and DeepSWE v1.1 (66.9), with fewer output tokens per task than several closed frontier models.

Point Claude Code, Codex CLI, or ZCode at https://t.co/TZKwd7fJif and set model="GLM-5.3".

Read more: https://t.co/Xkh4MRklFB

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include GLM-5.3, Zai_org, SambaCloud, 开源模型.

Source attribution: curated from SambaNova (@SambaNovaAI).