GLM-5.3 Hits the API at $1.40/$4.40 per Million Tokens

Z.ai logoZ.aiFYIAugust 20, 2026Pricing
What happened
GLM-5.3 is live on Z.ai's API at $1.40 input and $4.40 output per million tokens, unchanged from GLM-5.2.
Why it matters
Independent AA scoring puts it at 60 — level with Kimi K3, three points behind Claude Opus 5, with the open-weights tie taking effect once GLM-5.3's weights ship.
What to do
Model GLM-5.3's ~20% verbosity bump into your cost estimate — and hold off standardizing until the open weights drop.

GLM-5.3 is now callable on Z.ai's API at $1.40 per million input and $4.40 per million output tokens — unchanged from GLM-5.2 — and independent scoring puts it at 60 on the Artificial Analysis Intelligence Index, tying Kimi K3's score of 60 — the mark GLM-5.3 will match in the open-weights class once its weights ship. The pricing gap we flagged at launch is closed.

What happened

Z.ai opened GLM-5.3 to API traffic at the same rate as its predecessor, with cached input at $0.26 per million tokens (VentureBeat, 2026).

RatePrice per 1M tokens
Input$1.40
Output$4.40
Cached input$0.26

Artificial Analysis evaluated it at 60 on the Intelligence Index — level with Kimi K3 and three points behind Claude Opus 5, the closed-frontier leader (Unite.AI, 2026).

The catch is token efficiency. AA measured GLM-5.3 at roughly 18,700 output tokens per task, about 20% higher than GLM-5.2 — which lifts its real cost per Intelligence Index task to about $0.68 versus GLM-5.2's $0.44, still 19% cheaper than Kimi K3 per task (Artificial Analysis, 2026).

Why it matters

GLM-5.3 closes its last launch gap: published per-token pricing. Combined with a top-tier intelligence score that lands in the open-weights class once the weights ship, that puts it squarely on the Pareto frontier for teams choosing between Chinese open-weight models and the closed frontier.

The open weights are still held roughly two weeks for safety hardening — so our stance on GLM-5.3 stays conditional: reach for it on defensive-security and agentic workloads now, but hold off standardizing until the weights drop.

What changes for you

If you benchmarked GLM-5.3 and held off for pricing, the number is now live. Model two things into your decision:

  • Token efficiency, not just per-token rates. A ~20% verbosity bump means flat per-token rates are not flat per-task costs — $0.68 per index task versus $0.44 for GLM-5.2.
  • Wait for the weights. Self-hosting is still roughly two weeks out; don't lock your architecture until they ship.

FAQ

Did GLM-5.3's price change from GLM-5.2? No — it ships at the same $1.40/$4.40 per million tokens with $0.26 cached input. The news is that the rate is now published, not that it moved.

Is GLM-5.3 cheaper than Kimi K3? Per task, yes — about 19% cheaper, even though both score 60 on the Intelligence Index. The margin narrows because GLM-5.3 is more verbose per task.

Are the open weights available? Not yet. Z.ai is holding them roughly two weeks for safety hardening.

What to do

  1. 1 Model the ~20% verbosity bump into your per-task cost estimate before comparing GLM-5.3 to Kimi K3.
  2. 2 Hold off standardizing on GLM-5.3 until the open weights ship (~2 weeks out).

Affected tools & models

Never need to catch up again

The weekly delta — only verdict changes and act-now items. No digest filler.

By subscribing you agree to our Privacy Policy. Unsubscribe anytime.