GLM-5.3 open weights: Zhipu's 753B MoE coding model lands on Hugging Face (MIT)
Zhipu released open MIT-licensed weights for GLM-5.3, its 753B-parameter MoE coding/agentic model with 1M context - open-source SOTA on Terminal Bench 3.0, Agents' Last Exam (CLI), and CyberGym.
Why it matters
Post-training-only gains on the same GLM-5.2 base: a 6x jump on Terminal Bench 3.0 (4.6 to 28.3) plus emergent cyber capability (84.5 CyberGym), now freely self-hostable under MIT after the August API launch.
What you could build with it
A startup building dev tooling could self-host GLM-5.3 as the engine of a low-cost agentic coding product: MIT weights plus vLLM/SGLang support make it deployable on owned GPUs at a fraction of frontier API prices, with a 1M context window for whole-repo work.
Does it hold up?
Strong third-party signal for the API version (Bedrock availability, independent cost writeups); the open weights only landed Oct 14, so self-hosted usability reports are still to come.
Built with GLM-5.3 open weights
- GLM 5.3: Frontier Coding at 1/5 the Costtokenkarma.app · Independent cost analysis: GLM-5.3 matches frontier coding scores at roughly one-fifth the API price of GPT-5.6-class models, with a 100% pass rate on a 28-task real-world harness.
- GLM 5.3 on Amazon Bedrock for Coding and Cybersecuritythechenabtimes.com · AWS added GLM 5.3 to Bedrock on Oct 6, giving enterprise teams managed API access for coding and long-horizon agentic workloads.
- zai-org/GLM-5github · 7,348 ★ · Official GLM-5 series repo: deployment configs, inference examples, and post-training notes for the model family (verified via GitHub API).
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71NVIDIA Nemotron-pretraining-Indic-v1: 2.8T-token open Indic pretraining corpusNVIDIA with IIT Madras's AI4Bharat released a ~2.8 trillion-token pretraining corpus covering 22 Indian languages plus…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.