· CostMyTokens · Analysis · 6 min read

GLM-5.3 API Pricing: $1.40/$4.40, With Flash at $0.15/$0.50 — Full Comparison

GLM-5.3 API pricing: $1.40/$4.40 flagship, Flash $0.15/$0.50, FlashX $0.37/$1.25 — all with 1M context. Official Z.ai rates vs DeepSeek, Kimi and closed flagships.

Quick answer: Z.ai’s GLM-5.3 API costs $1.40 per million input tokens and $4.40 per million output tokens, with cached input at $0.26. The two cheaper tiers are GLM-5.3-Flash at $0.15/$0.50 (cached $0.03) and GLM-5.3-FlashX at $0.37/$1.25 (cached $0.075). All three share a 1M-token context window. There is no Batch tier — every request bills at these standard rates. Prices in US dollars, verified against Z.ai’s official pricing page on November 3, 2026.

Disclosure: this article contains no affiliate links. Where CostMyTokens does use affiliate links elsewhere, they never affect order — every table is sorted by price only. Read our disclosure.

GLM-5.3 prices at a glance

ModelInputCached inputOutputContextStandard workload*
GLM-5.3$1.40$0.26$4.401M$20.00/mo
GLM-5.3-FlashX$0.37$0.075$1.251M$5.46/mo
GLM-5.3-Flash$0.15$0.03$0.501M$2.20/mo

* Our standard workload: 10M tokens a month at an 8:2 input/output split, no caching. Full per-model tables and live calculators are on the model pages for GLM-5.3, GLM-5.3-Flash and GLM-5.3-FlashX, and the complete Z.ai lineup is on the Z.ai provider page.

Two reading notes on the official table. First, cached input storage is currently listed as “Limited-time Free” — you pay the cached-input rate when a cached prompt is read, but storing the cache costs nothing for now; because Z.ai labels it limited-time, treat it as a promotion that can end, not a permanent term. Second, the flagship price did not move with the generation: GLM-5.2 and GLM-5.1 sit on the same official page at the same $1.40/$4.40, so 5.3 is a capability upgrade at a held price.

How the three tiers differ

All three are current-generation models with the same 1M context and 128K maximum output; what changes is speed, modality and price:

  • GLM-5.3 ($1.40/$4.40) — the flagship. Text-only, with always-on reasoning. The tier for quality-first work: complex reasoning, long documents, agentic tasks where output quality dominates the bill.
  • GLM-5.3-FlashX ($0.37/$1.25) — the faster-serving Flash tier. Multimodal (image, video and file inputs) at roughly a quarter of the flagship’s input price, for latency-sensitive production traffic.
  • GLM-5.3-Flash ($0.15/$0.50) — the volume tier. Also multimodal, and priced for classification, extraction, summarisation and other high-call-volume work where the flagship’s reasoning is unnecessary.

The spread is unusually wide: the flagship costs 9.3x the Flash tier on input and 8.8x on output, so tier choice matters more to the monthly bill than any caching trick. Route by task, not by habit.

What the tiers cost on real workloads

Beyond the standard workload in the table above, a blended mix of 1M input + 250K output tokens (a typical chat or assistant pattern) costs:

  • GLM-5.3: $2.50
  • GLM-5.3-FlashX: $0.68
  • GLM-5.3-Flash: $0.28

Caching compresses these further. Cached input bills at roughly one fifth of the standard input rate on every tier ($0.26 vs $1.40 on the flagship; $0.03 vs $0.15 on Flash), so a workload that reuses long system prompts or documents — support bots, RAG pipelines, document review — can cut its input spend by up to ~80% on the cached share. Our API cost calculator applies the cached rate to whatever share of input you mark as reused.

One add-on to budget: Z.ai’s built-in Web Search tool costs $0.01 per use when your application calls it. It is opt-in and billed per call, separate from token charges.

GLM-5.3 vs the rival flagships

Against the other commercial flagships we track at verified official rates:

ModelInputOutputStandard workload*
GLM-5.3-Flash$0.15$0.50$2.20
DeepSeek V4.1 Flash (off-peak)$0.15$0.60$2.40
Kimi K2.6$0.95$4.00$15.60
DeepSeek V4 Pro (peak)$1.32$3.96$18.48
GLM-5.3$1.40$4.40$20.00
Grok 4.7$2.00$6.00$28.00
GPT-6.1 Sol / Claude Sonnet 5.5$2.00$10.00$36.00

* 10M tokens/month, 8:2 input/output, standard rates, no caching. DeepSeek prices time-of-day: the Pro row is peak (off-peak is half, $9.24 on this workload); full detail in our DeepSeek API pricing guide.

The positioning is precise. GLM-5.3 prices its flagship just above DeepSeek’s peak rate and well below the $2-input Western flagships, with a noticeably cheaper output rate than the $10-output tier — output-heavy workloads (long generations, code, reports) feel that gap most. At the bottom, GLM-5.3-Flash ties DeepSeek’s off-peak Flash input price and undercuts its output ($0.50 vs $0.60), without the time-of-day condition — the same $0.15/$0.50 applies around the clock. On our cheapest LLM API ranking, that puts the Flash tier in the first rank of production-usable models, and even the flagship’s $20.00 standard workload lands mid-table rather than with the premium flagships.

Which tier should you pick?

  • Start with Flash for volume work — classification, extraction, tagging, first-pass summarisation. At $2.20 per standard workload it is cheap enough that quality checks, not cost, should decide whether you move up.
  • Pick FlashX when the same work is user-facing and latency matters, or when you want one tier to cover both text and image/video inputs at production speed.
  • Pay for GLM-5.3 when reasoning quality is the product: hard analysis, long-context document work, agents whose mistakes cost more than tokens. Even then, route the easy calls down — a flagship-only setup pays 9x for input that a Flash tier would have handled.

FAQ

How much does the GLM-5.3 API cost? $1.40 per million input tokens, $0.26 per million cached input tokens and $4.40 per million output tokens. The Flash tier costs $0.15/$0.50 and FlashX $0.37/$1.25, with cached input at $0.03 and $0.075 respectively.

Is there a Batch or off-peak discount on GLM-5.3? No. The official pricing table has no Batch column for the GLM-5.3 family, and Z.ai does not run time-of-day pricing — the listed rates apply around the clock. The only standing discount is cached input at roughly one fifth of the standard input rate.

What does cached input storage cost? Z.ai currently lists cached input storage as limited-time free. Reading cached input is billed at the cached-input rate; because the free storage is labelled limited-time, it may become chargeable later.

How does GLM-5.3 compare with DeepSeek on price? GLM-5.3 ($1.40/$4.40) sits just above DeepSeek V4 Pro’s peak rate ($1.32/$3.96) and well above its off-peak rate, while GLM-5.3-Flash ($0.15/$0.50) matches DeepSeek Flash’s off-peak input price with cheaper output — and no time-of-day condition. See our DeepSeek API pricing guide for the full peak/off-peak tables.

Does GLM-5.3 have a free tier? The GLM-5.3 family itself is paid at the rates above. Z.ai’s official page lists some older models (GLM-4.7-Flash, GLM-4.5-Flash) as free, but those are previous-generation models, not the 5.3 family covered here.

What does the Web Search tool cost? $0.01 per use, billed per call when your application invokes Z.ai’s built-in web search, on top of token charges.

Browse every official Z.ai model on the Z.ai provider page, price your own traffic on the GLM-5.3, GLM-5.3-Flash or GLM-5.3-FlashX model pages, and compare with DeepSeek API pricing, Hunyuan Hy4 API pricing, Kimi API pricing and the cheapest LLM API in 2026.

Prices and plan details verified against Z.ai’s official pricing page (docs.z.ai/guides/overview/pricing) on November 3, 2026. Z.ai states prices in USD per million tokens unless noted. For the full Z.ai lineup, see the Z.ai provider page.

Share:
Back to Blog

Related Posts

View All Posts »