The model race now has a speed tier. zhipu's glm-5.3-flashx runs 18b active params out of 320b on…
the model race now has a speed tier. zhipu's glm-5.3-flashx runs 18b active params out of 320b on hybrid sparse-plus-linear attention, hits 200 tokens a second, and cuts kv cache 4.4x.
inference speed is the release axis now, not leaderboard points.
Context
Z.ai's post of 26 August 2026 describes GLM-5.3-Flash as 320B total and 18B active parameters on hybrid sparse and linear attention, with 3.0x less attention compute and a 4.4x smaller KV cache than GLM-5.3, a KV cache it says is still slightly larger than Kimi-K3 and DeepSeek-V4-Flash, with weights on Hugging Face. Z.ai claims it outperforms GLM-5.2 at one-tenth the price and approaches Claude Opus 4.8 on coding and agentic benchmarks, which are vendor claims.
The architecture facts are first-party for GLM-5.3-Flash, not for glm-5.3-flashx. The 4.4x is against the larger GLM-5.3, not against Flash or FlashX. The 200 tokens a second and the FlashX designation were not on the Z.ai pages read, and the pricing page copy has no GLM-5.3 FlashX row. The earlier FlashX note drew on secondary reports of a launch on 18 September 2026 at a vendor-peak 200 tokens a second, which stay as positive secondary evidence and are not reconciled with a first-party FlashX page. The Flash post leads with benchmarks and price as well as efficiency. That inference speed is the release axis is the author's thesis.
Related work
- Earlier note on GLM-5.3 FlashX ↗FlashX versus Flash.
Watch next
- A first-party FlashX page with its spec and tokens-per-second configuration, and independent throughput tests.
Sources
- GLM-5.3-Flash (Z.ai, 26 Aug 2026)z.ai
- GLM-5.3-Flash guide (Z.ai docs)docs.z.ai
- Pricing (Z.ai docs)docs.z.ai
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 04:04 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →