GLM-5.3 Flash: 320B MoE Model Matches Opus at 1/10th Cost
kimmonismus · x · 2026-08-26
Zai's upcoming GLM-5.3 Flash (Ox Alpha) is a 320B MoE model activating only 18B parameters. It scores 84.3 on Terminal-Bench 2.1, nearing Claude Opus 4.8, and outperforms Opus on DeepSWE and AutomationBench. It offers higher efficiency than GLM-5.2 at one-tenth the serving cost, features open-source MIT licensing, native multimodality, and a 1M context window.
Related event: Leak: Ox Alpha Is GLM-5.3 Flash, 320B MoE at One-Tenth the Price(4 posts)→
More from Models
- Baseten Loops Adds Support for RL and Fine-tuning GLM-5.3-Flash — baseten · 2026-08-26
- GLM-5.3-Flash Now Available on Nous Portal — NousResearch · 2026-08-26
- ZAI's 0xAlpha Cuts Inference Costs 10x by Adopting Peer Innovations — zephyr_z9 · 2026-08-26
- GLM-5.3 Flash Arrives on OpenRouter, Succeeding Ox Alpha — Teknium · 2026-08-26
- Qwen Tech Report Praised: Meticulous Detail, N-gram Method Could Become Standard — code_star · 2026-08-26
- Zai Releases GLM-5.3-Flash: 320B MoE with Hybrid Attention and 1M Context — TheZachMueller · 2026-08-26