Top AI Product

Z.ai shipped GLM-5.3 on August 14. Same 743B base as GLM-5.2 — every gain came from scaled-up post-training. Coding jumped ~50% in internal evals, and Terminal-Bench 3.0 went from 4.6 to 28.3. That’s a 6x move without touching the base model.

The capability nobody trained for

Z.ai added vulnerability-discovery environments hoping the model would get better at spotting single bugs. What came out plans complete multi-stage exploit chains. It hit 84.5% on CyberGym, edging past Mythos 5 and GPT-5.6 Sol. First time an open-weight model tops closed frontier models on offensive security — and it was emergent, not the goal. Since GLM-5.2, Z.ai models have found 2,436 vulns across 269 open-source projects, 1,097 critical or high.

Using it today

It’s a text model you hit through the Z.ai API, priced like 5.2: $1.40 per million input, $4.40 output, $0.26 cached. Thinking is mandatory now — low, high, max, no off switch. The obvious use case is long-horizon agentic coding, where it lands near Claude Fable 5 at a fraction of the cost. Weights drop in two weeks.

You Might Also Like

• Qwen3 8 max Goes ga Alibabas 2 4t Model Beats Claude Fable 5 on Terminal Bench Weights Open Next Week
• Glm 5 Just Dropped and its the Open Source Model Nobody saw Coming
• Ornith 1 0 Deepreinforce Self Scaffolding Coding Models Open Weights That Write Their own rl Scaffold 397b Hits 82 4 on swe Bench
• Zcode Zhipu z ai glm 5 2 Coding Agent hit hns 1 Spot Chinas Open Answer to Claude Code
• Castform x Neon a 4b Open Source Model Matches gpt 5 6 sol on Retrieval at 1 100 the Cost