Model Releases
The chart says GLM-5.1 scored 54.9 on coding benchmarks. Three points behind Claude Opus 4.6. Interesting but not the story. The story is wh…
The chart says GLM-5.1 scored 54.9 on coding benchmarks. Three points behind Claude Opus 4.6. Interesting but not the story. The story is what trained it. Zero Nvidia GPUs. 100,000 Huawei Ascend 910B
The chart says GLM-5.1 scored 54.9 on coding benchmarks. Three points behind Claude Opus 4.6. Interesting but not the story. The story is what trained it. Zero Nvidia GPUs. 100,000 Huawei Ascend 910B chips. Every parameter. Every one of the 28.5 trillion training tokens. Domestic Chinese silicon from start to finish. The Ascend 910B delivers roughly 320 TFLOPS of FP16. An H100 does 989. Each chip is about a third as powerful as what OpenAI and Anthropic use. http://Z.ai compensated with sheer volume and custom distributed training infrastructure they had to build from scratch because no existing framework was designed for Ascend at this scale. US chip export controls were specifically designed to prevent this. The Commerce Department put Zhipu AI on the entity list. Cut off H100 and A100 access. Closed the H800 loophole. The explicit theory: restrict hardware, keep Chinese labs a generation behind. GLM-5.1 is 3.1 points behind Opus 4.6 on coding. The distance between "a generation behind" and "within rounding error" is the entire AI story of 2026. http://Z.ai IPO'd in Hong Kong in January at $6.5B. Stock surged 500%+ within weeks. API priced at ~$1/M input tokens vs $5 for Opus 4.6. Model is MIT-licensed. Anyone on Earth can download, modify, and deploy it commercially. The sanctions created a competitor with zero Western supply chain dependency and every incentive to undercut on price. Introducing GLM-5.1: The Next Level of Open Source - Top-Tier Performance: #1 in open source and #3 globally across SWE-Bench Pro, Terminal-Bench, and NL2Repo. - Built for Long-Horizon Tasks: Runs autonomously for 8 hours, refining strategies through thousands of iterations. Blog…
Related
- http://Z.ai releases GLM-5.1, a 754B-parameter model that it says outperforms GPT-5.4 and Claude Opus 4.6 on SWE-bench Pro, available under …
- INCREDIBLE GLM-5.1 weights are now opensource > i’ve had early access to the weights for the past few days > and yeah… this one matters a lo…
- To run GLM-5.1 locally (744B params, 40B active MoE), full precision needs ~1.65TB disk + enterprise hardware like 8x H200/B200 GPUs. Minimu…
- A lot of you are asking about the :cloud tag. Here's the deal: GLM-5.1 is a 744B parameter beast. To hit that 54.9 benchmark score without t…
- 【速報】中国のAI企業http://Z.ai（旧Zhipu AI）が最新AIモデル「GLM-5.1」をリリースしました🚀 コーディング性能を測る「SWE-Bench Pro」で58.4%を記録し、オープンソース（誰でも無料で使える形）モデルとして世界1位。全体でも3位（GPT-…
Source: Zhipu AI (X) | 2026-04-07