GLM-5.3 runs half price on Vercel's AI Gateway through Tuesday
It's the DigitalOcean route doing it, and it snaps back Sept 8 — so pin the provider now. GLM-5.3 leads open models on Terminal-Bench, at 1M context.
Copy markdown
One day left on the 50% cut
GLM-5.3 runs at half its list rate on Vercel AI Gateway through Tuesday, Sept 8, via the DigitalOcean route — then it's back to full price. At a ~$1.40/$4.40 per-1M list (in/out), that's roughly $0.70/$2.20 while the window holds.
Pin the provider, or lose the price
Two ways in: the promo id `zai/glm-5.3-promo-50` routes only to DigitalOcean and stops serving the moment the offer ends, or use `zai/glm-5.3` with `providerOptions.gateway.order: ['digitalocean']` to keep hitting that provider afterward. One locks today's price; the other keeps the route.
Why it's worth pinning
GLM-5.3 is a ~744B-class MoE (about 40B active), served at up to 1M context and 128K max output on the gateway. It ranks first among open models on Terminal-Bench 3.0 at 28.3 — up from GLM-5.2's 4.6 — so it's real agentic-coding muscle at a fraction of frontier prices.
The rest of the family is routable too
GLM 5.3 Flash and GLM 5.3 Fast are both live on Vercel AI Gateway alongside the base model now, so you can drop to a cheaper or lower-latency tier per call without leaving the gateway or rewriting your client.
Elsewhere: DeepSeek's first native vision model ships MIT
If you missed it last week, DeepSeek's V4-Flash-Vision-Exp weights are out under MIT (~168GB in FP8/FP4, 1M context, 284B total / 13B active). It reads charts, screenshots and docs and edges Opus-4.8 on Agents' Last Exam — weekend fuel for an open multimodal agent.