GLM-5.3-Flash doubles on September 9
Live grant $0.075 in / $0.25 out becomes list $0.15 in / $0.50 out, per million tokens
The launch discount ends on September ninth. Then the bill doubles. GLM-5.3-Flash is the model on that clock. GLM-5.3-Flash is Z.ai’s cheaper, faster sibling of GLM-5.3: three hundred and twenty billion in the file, eighteen billion awake per word, a million-word window. Z.ai’s live grant is seven and a half cents per million input tokens, the text you send, one and a half cents per million cached input tokens, and twenty-five cents per million output tokens, the text the model writes back. The crossed-out number is the real menu price. The live number is a fifty percent off grant with an end date. List is fifteen cents in, three cents cached, and fifty cents out, per million tokens. Z.ai times that grant at midnight on September ninth, Singapore time. OpenRouter times the same instant at four in the afternoon UTC. Your API bill can double on a calendar date.
GLM-5.3-Flash lists at $0.15 in and $0.50 out
GLM-5.3-Flash lists at $0.15 in and $0.50 out, per million tokens
Released 2026-08-26 · Z.ai · MIT from the LICENSE file
You can hand GLM-5.3-Flash about a million tokens at once. Z.ai caps the reply at a hundred and twenty-eight thousand tokens. Z.ai lists GLM-5.3-Flash at fifteen cents per million input tokens and fifty cents per million output tokens. Artificial Analysis clocks GLM-5.3-Flash on Z.ai’s API at forty-seven point six tokens a second. That is how fast the words come out. The wait before the first word is one point six four seconds. The LICENSE file is MIT, copyright two thousand twenty-six, Z.AI. MIT lets you use the weights, including commercially, with almost no extra gates.
Intelligence Index v4.2 scores 46 at $0.18 per task
Artificial Analysis scores GLM-5.3-Flash 46 on Intelligence Index v4.2, at $0.18 per task
Fourth of 112 large models you can download · live number on v4.2
v4.2 is a different mix of tests than v4.1.1
Artificial Analysis scores GLM-5.3-Flash forty-six on Intelligence Index version 4.2. The Intelligence Index is a single score Artificial Analysis builds by running the same tests on every model itself. Cost per Index task is what Artificial Analysis billed to finish one of those tests. Artificial Analysis billed eighteen cents per Intelligence Index task. On August twenty-sixth, launch day, Artificial Analysis scored GLM-5.3-Flash fifty-seven on Intelligence Index version 4.1.1, at nine cents per Index task. Version 4.2 is a different mix of tests than version 4.1.1.
Artificial Analysis’s $0.10 blend is built at list
A budget past September 9 plans ten cents per million tokens
Blended token rate at list: $0.10 /Mtok at 7:2:1 · Task rate: $0.18 per Intelligence Index task
Z.ai is selling the same eighteen-billion-awake model at half list until a printed instant. After midnight Singapore time on September ninth, Z.ai’s list is fifteen cents in, three cents cached, and fifty cents out, per million tokens. Artificial Analysis’s ten cents blended per million tokens is built at that list. The task rate is a different number: eighteen cents per Index task, also built at list. A blended price is one dollar figure standing in for a mixed bill: seven parts prompt text the model has already seen, two parts new prompt, one part what it writes back. The mix uses seven parts cached at three cents, two parts new input at fifteen cents, and one part output at fifty cents. That mix is ten point one cents. Artificial Analysis rounds the mix to ten cents per million tokens. At the live grant the same mix uses one and a half cents per million cached tokens, seven and a half cents per million new input tokens, and twenty-five cents per million output tokens. That mix is about five cents per million tokens. A budget that survives past September ninth has to plan the ten cents per million tokens. Z.ai still prints cached input storage as Limited-time Free on the same page. That storage grant has no September ninth date.
Four OpenRouter hosts still show 50% off
Four OpenRouter hosts still show 50% off; sixteen already bill list $0.15 in / $0.50 out
Z.ai’s API doubles after the clock. CoreWeave, Fireworks, and Together already bill list.
24 hosts on OpenRouter · banner: through 9 September 2026 at 16:00 UTC
Z.ai wrote the clock. OpenRouter’s banner matches four in the afternoon UTC on September ninth. OpenRouter lists twenty-four hosts for GLM-5.3-Flash today. Four hosts still show fifty percent off: Z.ai, Novita, DeepInfra, and GMICloud, at seven and a half cents in and twenty-five cents out, per million tokens. Sixteen hosts already bill list: fifteen cents in and fifty cents out, per million tokens. CoreWeave, Fireworks, and Together sit in that already-list group. CoreWeave’s cached input is five cents per million tokens. Z.ai’s cached list is three cents per million tokens. Wafer prints a different sticker: ten cents in and thirty-five cents out, per million tokens, with no promo badge. Modal takes sixty-seven percent off a forty-five cent in and a dollar fifty out list, and lands at fifteen cents in and fifty cents out. A budget that routes through Z.ai’s API pays double after that clock. A budget that routes through CoreWeave, Fireworks, or Together already pays list today.
GLM-5.3-Flash at 46 index points and $0.10 blended
GLM-5.3-Flash sits at 46 index points and $0.10 blended at list
Intelligence Index v4.2 against blended $/Mtok at list. Qwen 46 is an AA estimate.
GLM-5.3-Flash sits at forty-six index points, ten cents blended per million tokens, and eighteen cents per Index task, all at list. Qwen3.8-Flash-Next sits at the same forty-six index points, marked as an Artificial Analysis estimate, at nine cents blended per million tokens. Gemini 3.8 Flash on high scores forty-seven index points at fifty-eight cents blended per million tokens. The task rate for Gemini 3.8 Flash on high is a different number: seventy-four cents per Index task. GLM-5.3 on max scores forty-nine index points at ninety cents blended per million tokens, and a dollar twenty-six per Index task.
Past September 9 on Z.ai, plan $0.15 in and $0.50 out
This week’s promo is half that. Some hosts already bill list.
If the bill has to survive past September ninth on Z.ai, plan fifteen cents in and fifty cents out per million tokens, Index forty-six, eighteen cents per Index task, forty-seven point six tokens a second. If you are still on the promo this week, the live invoice is seven and a half cents in and twenty-five cents out, per million tokens, and the clock is midnight Singapore time on September ninth. If you pick a host by sticker, OpenRouter already prints the split. Wafer bills ten cents in and thirty-five cents out, per million tokens. The MIT weights stay MIT if you run the files yourself. The doubling is an API event.
At midnight Singapore time the strikethrough becomes the list
At midnight Singapore time on September 9 the strikethrough becomes the list
Same 18B-awake model. New invoice.
The promo bars from the open become Z.ai’s list at midnight Singapore time on September ninth. The model stays the same. The invoice doubles. After four in the afternoon UTC that day, the next reading is whether Novita, DeepInfra, and GMICloud follow Z.ai to list.
