THE 60-SECOND VERSION
● GLM-5.3 weights expected today after a two-week hold. Licence still unstated.
● Ox Alpha was GLM-5.3-Flash — 320B, MIT, and served entirely on Chinese AI chips.
● The bigger pattern: two labs gated cyber-capable models this month.
● 3 days to the 31 August double deadline.
GLM-5.3 weights land today
The zai-org page on Hugging Face lists 28 August for GLM-5.3 open weights, roughly two weeks after the 14 August launch. Z.ai held them for what it calls its most extensive risk review to date, citing cyber capability that emerged faster than expected.
That break matters because GLM-5.2 shipped MIT-licensed weights within days of launch, and the immediacy was the whole proposition. The flagship licence for 5.3 has still not been stated. Until the weights appear under zai-org, any GLM-5.3 upload from anyone else is not the model.
Benchmarks from Z.ai's own testing: Terminal-Bench 3.0 from 4.6 to 28.3, DeepSWE v1.1 from 46.2 to 66.9, and CyberGym at 84.5 percent. All gains from post-training on the same base as GLM-5.2, with no new pre-training.
Full breakdown: the benchmarks, the licence question, and the Flash release →
And the pattern underneath it
Z.ai is not alone. OpenAI gated its own cyber-capable model within the same fortnight, Anthropic disclosed a frontier model it will not ship, and Alibaba released Qwen 3.8-Max under a custom licence rather than Apache.
Four labs, four jurisdictions, one direction — launched and downloadable are becoming separate events. If your roadmap assumed frontier weights arriving on a predictable schedule, that now needs a hedge.
Both sides of the argument, and what to change →
Three days to the double deadline
| Date |
What happens |
| Aug 31 |
Claude Sonnet 5 goes $2 to $3 per million input, $10 to $15 output, plus a tokenizer change adding 10 to 35 percent tokens on code |
| Aug 31 |
GPT-5.4 and GPT-5.4 mini leave Codex for ChatGPT sign-in users. Still available via API key |
| Aug 31 |
kimi-k2.5 and moonshot-v1 sunset, migrate to kimi-k3 |
| Early-mid Sept |
Grok 4.7 window per Musk. Still no model ID at docs.x.ai |
| Oct 1 |
OpenAI vs Apple hearing |
| Oct 24 |
deepseek-chat and deepseek-reasoner deprecated |
| ~Nov 21 |
GPT-5.6 Sol promotional pricing ends, $4/$20 back to $5/$30 |
Monday is a repricing and a migration at once if you run coding workloads across both Anthropic and OpenAI. The tokenizer change is the part being missed — same file, more tokens, higher rate.
Still open
FAQ
When do GLM-5.3 weights release?
The zai-org page on Hugging Face lists 28 August 2026, about two weeks after the 14 August launch. A listed date can still move.
What licence will GLM-5.3 use?
Not stated. GLM-5.2 was MIT and GLM-5.3-Flash was released under MIT on 26 August, but the flagship has not been confirmed.
What was Ox Alpha?
GLM-5.3-Flash, running anonymously on OpenRouter before Z.ai confirmed it on 26 August. 320B total with 18B active, MIT licensed, and served entirely on Chinese AI chips.
What changes on 31 August?
Claude Sonnet 5 rises to $3 in and $15 out per million with a tokenizer change alongside, and GPT-5.4 and GPT-5.4 mini leave Codex for ChatGPT sign-in users. Both remain available via API.