--- title: "GLM-5.3-Flash costs $1.10 per 1,000 8k/2k calls until September 9" description: "$1.10 per 1,000 calls doubles to $2.20 after GLM-5.3-Flash's promotion ends September 9. Prices verified 2026-08-27." canonical: https://vynaris.com/blog/glm-53-flash-110-per-1000-calls-promo-expiry author: "Dhrumil Barot" author_bio: "Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers." published: 2026-08-27 updated: 2026-08-27 --- # GLM-5.3-Flash costs $1.10 per 1,000 8k/2k calls until September 9 Published 2026-08-27 · By [Dhrumil Barot](https://vynaris.com/about#founder), Founder $1.10 per 1,000 calls doubles to $2.20 after GLM-5.3-Flash's promotion ends September 9. Prices verified 2026-08-27. [GLM-5.3-Flash](https://vynaris.com/models#glm-5-3-flash) costs $1.10 per 1,000 calls with 8,000 uncached input and 2,000 output tokens. That bill doubles to $2.20 when its 50% launch promotion ends at 2026-09-09 16:00 UTC. The discount is useful; the deadline is the routing decision. Prices verified 2026-08-27. ## TL;DR - The promotional rates are $0.075 per 1M input tokens, $0.015 per 1M cached input tokens and $0.25 per 1M output tokens. - The same rates return to $0.15, $0.03 and $0.50 after the promotion. A fixed 8k-input/2k-output call moves from $0.0011 to $0.0022. - Z.AI reports 63.4 on DeepSWE v1.1 and 48.8 on AutomationBench. Treat those as test candidates, not proof for your workload. ## Verdict table | Decision receipt | Promotional rate | After 2026-09-09 16:00 UTC | Change | | --- | --- | --- | --- | | Input per 1M tokens | $0.075 | $0.15 | 2x | | Cached input per 1M tokens | $0.015 | $0.03 | 2x | | Output per 1M tokens | $0.25 | $0.50 | 2x | | 1,000 calls, 8k uncached input + 2k output each | $1.10 | $2.20 | +$1.10 | | 1,000 calls, all 8k input tokens cached + 2k output each | $0.62 | $1.24 | +$0.62 | Buyers should benchmark now, then budget at the list rate. A temporary sticker is a poor permanent forecast. ## The $1.10 bill, line by line Z.AI's live [pricing page](https://docs.z.ai/guides/overview/pricing) lists the promotion and its end time. It states 24:00 on September 9 in Singapore time. That equals 2026-09-09 16:00 UTC. For 1,000 uncached calls, input usage is 8 million tokens. Output usage is 2 million. The 8k input is a disclosed workload shape, not a claim about the model's [context window](https://vynaris.com/glossary/context-window). The promotional bill is: `8M × $0.075/M + 2M × $0.25/M = $0.60 + $0.50 = $1.10` After the promotion, the same workload becomes: `8M × $0.15/M + 2M × $0.50/M = $1.20 + $1.00 = $2.20` That is $0.0011 per call now and $0.0022 later. The absolute increase is $1.10 per 1,000 calls. At one million calls, it is $1,100. The cached case assumes every input token qualifies for the listed [prompt caching](https://vynaris.com/glossary/prompt-caching) rate. It is a boundary, not a forecast. Real cache hits depend on stable prefixes, request order and provider rules. Use [the calculator](https://vynaris.com/calculator?in=8000&out=2000) with your own [input token cost](https://vynaris.com/glossary/category/routing-and-cost#input-token-cost), output length and call volume. ## The quality receipts earn a test, not a migration The [launch report](https://z.ai/blog/glm-5.3-flash) pairs the low rate with public benchmark results. Z.AI reports an Artificial Analysis Intelligence Index v4.1.1 score of 57 at $0.045 per task under the discount. It also reports 63.4 versus GLM-5.2's 46.2 on DeepSWE v1.1. That is a 17.2-point gap. On AutomationBench, it reports 48.8 versus 26.2, a 22.6-point gap. Those scores address a real objection: the cheaper model is not presented as price-only. They still do not price our disclosed 8k/2k shape. The $0.045 benchmark task uses its own token mix. We should not reverse-engineer that workload from one total. The provider published these comparisons. The Artificial Analysis result is third-party, while several other claims appear in the provider's launch report. Neither predicts your acceptance rate, latency or tool-call accuracy. Our [GLM-5.3 quality-adjusted cost analysis](https://vynaris.com/blog/glm-53-28-task-cost-vs-gpt-55-quality-adjusted-bill) shows the safer pattern. Divide spend by passed tasks on your own eval. Our earlier [GLM-5.2 model comparison](https://vynaris.com/blog/glm-52-vs-opus-48-vs-gpt55-cost-per-task) supplies a broader sticker-price baseline. ## What should change before September 9 First, tag GLM-5.3-Flash spend separately. A blended provider invoice can hide the step-up. Second, run a representative sample before the deadline. Measure [cost per task](https://vynaris.com/glossary/category/routing-and-cost#cost-per-query), accepted outcomes and tail latency. A leaderboard score cannot set your production threshold. Third, set two budget alerts. The first uses today's $0.075/$0.25 rates. The second uses $0.15/$0.50. If the second alert breaks the unit economics, the workload needs a smaller output, a higher cache-hit rate or a different route. Finally, keep the promotion end in UTC in code and procurement notes. “September 9” without a timezone is how an avoidable invoice becomes a postmortem. ## Honest tradeoff: the cheapest route may be operationally expensive Do not [route models](https://vynaris.com/glossary/model-routing) on this promotion alone. Vendor approval, data location and incident response can outweigh $1.10 per 1,000 calls. A failed coding-agent task can also consume reviewer time that dwarfs the API bill. The opposite mistake is ignoring a time-boxed test window. The promotion halves a controlled evaluation bill. Use it to collect evidence. Then make the production decision against list rates and a quality-adjusted denominator. ## Sources - [Z.AI developer pricing](https://docs.z.ai/guides/overview/pricing), rates and promotion deadline verified 2026-08-27. - [Z.AI GLM-5.3-Flash launch report](https://z.ai/blog/glm-5.3-flash), model name and benchmark receipts verified 2026-08-27. - [Hacker News launch discussion](https://news.ycombinator.com/item?id=49449507), 927 points and 464 comments verified 2026-08-27. - Reproducible arithmetic: `artifacts/glm-53-flash-110-per-1000-calls-promo-expiry-math.py`. ## About the author [Dhrumil Barot](https://vynaris.com/about#founder), Founder. Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers. [LinkedIn](https://www.linkedin.com/in/bdhrumil) · [X](https://x.com/fyidhrumil)