Z.AI Pricing: What GLM Models Actually Cost

Z.AI is the international brand of Chinese AI lab Zhipu AI. Its GLM models have emerged as a frequently cited, lower-cost alternative to Claude and GPT. But comparing costs isn't straightforward: Z.AI's pricing page combines per-token API rates, a subscription-based Coding Plan, and a limited free tier.

This page uses figures directly from Z.AI's pricing documentation, with the API rates verified on 2026-10-07. Every published price is sourced, and where public pricing isn't available, such as Coding Plan Pro and Max tiers or enterprise contracts, we make that clear and direct you to the live page rather than filling in the gaps.

Below, you'll find API rates for every available GLM tier, details on what the free tier includes, a high-level overview of the GLM Coding Plan, and what enterprise buyers can expect. The full subscription breakdown is available on our GLM Coding Plan page. We also compare Z.AI's pricing with Anthropic Claude, OpenAI, and DeepSeek.

Model prices change frequently, so always check the vendor's live pricing page before signing a contract or building a cost model.

Z.AI Pricing at a Glance

Z.AI sells access to its GLM models three ways. Understanding which one you need is the first cost decision, because the price-per-token math only applies to one of them.

• Pay-as-you-go API: per-1M-token pricing for input, output, and cached input. This is the standard developer path and the mode most cost comparisons refer to.
• GLM Coding Plan: a flat monthly subscription with prompt quotas, built around the GLM Coding tier. Starts at $18/month for the Lite plan. Suited to individual developers using the model inside coding IDEs.
• Free tier: GLM-4.7-Flash, GLM-4.5-Flash, and GLM-4.6V-Flash are priced at zero for input, output, and cached input on Z.AI's public API pricing page. They are genuine free models, not a trial credit.
• Enterprise / custom: volume commits and dedicated deployments are not publicly listed — contact Z.AI sales.

API Pricing by GLM Model

These are Z.AI's published per-1M-token API rates as listed on the official pricing page on 2026-10-07. All values are USD. Cached-input pricing applies when a prefix has been seen recently and Z.AI can reuse the KV cache.

• GLM-5.3 (current flagship, released 2026-08-17): $1.40 per 1M input, $4.40 per 1M output, $0.26 per 1M cached input.
• GLM-5.3-Flash: $0.15 per 1M input, $0.50 per 1M output, $0.03 per 1M cached input. Cheap, but not free.
• GLM-5.3-FlashX: $0.37 per 1M input, $1.25 per 1M output, $0.075 per 1M cached input.
• GLM-5.2 and GLM-5.1: $1.40 per 1M input, $4.40 per 1M output, $0.26 per 1M cached input, the same rates as GLM-5.3.
• GLM-5: $1.00 per 1M input, $3.20 per 1M output, $0.20 per 1M cached input.
• GLM-4.7: $0.60 per 1M input, $2.20 per 1M output, $0.11 per 1M cached input.
• GLM-4.6: $0.60 per 1M input tokens, $2.20 per 1M output tokens, $0.11 per 1M cached-input tokens.
• GLM-4.5: $0.60 per 1M input, $2.20 per 1M output, $0.11 per 1M cached input. Same public rates as GLM-4.6.
• GLM-4.5-Air (lightweight): $0.20 per 1M input, $1.10 per 1M output, $0.03 per 1M cached input.
• GLM-4.5-Flash: free — $0 for input, output, and cached input on the public pricing page.
• GLM-4.5V (vision): $0.60 per 1M input, $1.80 per 1M output, $0.11 per 1M cached input.
• GLM-4-32B-0414-128K: $0.10 per 1M input, $0.10 per 1M output.
• GLM-4.7-Flash and GLM-4.6V-Flash are also listed at $0, alongside GLM-4.5-Flash.
• The GLM-5 family now appears on the same public per-token table as the 4.x line. That was not true when this page was first written, so an older cost model built on "GLM-5 is Coding Plan only" needs redoing against the rates above.

GLM Coding Plan (Subscription)

The GLM Coding Plan is a flat monthly subscription that gives coding-tool users a prompt quota rather than per-token billing. It is aimed at developers using GLM inside IDEs and CLI coding assistants.

Z.AI's docs list three tiers — Lite, Pro, and Max — each with its own 5-hour and weekly prompt cap. The Lite plan starts at $18/month. Pro and Max monthly prices are not published on the docs overview page verified here; check z.ai/subscribe for the current tier prices.

For the full breakdown of quotas, model access, and how the Coding Plan compares to per-token API billing, see our dedicated /guides/glm-coding-plan-explained page.

• Lite: ~80 prompts per 5 hours, ~400 per week. Starts at $18/month.
• Pro: ~400 prompts per 5 hours, ~2,000 per week. Monthly price not publicly listed on the docs overview page — check z.ai/subscribe.
• Max: ~1,600 prompts per 5 hours, ~8,000 per week. Monthly price not publicly listed on the docs overview page — check z.ai/subscribe.
• Z.AI states the monthly quota is roughly equivalent to 15–30x the subscription fee in API-billing terms, so the plan is designed to be cheaper than per-token API for steady coding workloads.
• Advanced models (GLM-5.2, GLM-5-Turbo) consume quota at higher multiples during peak Beijing hours.

Free-Tier Limits

Three Z.AI models are free on the public API: GLM-4.7-Flash, GLM-4.5-Flash, and GLM-4.6V-Flash. GLM-5.3-Flash is not one of them, which is the single most common mix-up on Z.AI pricing, and the flagship GLM-5.3 is not free either. Z.AI gives free access by pricing whole models at zero rather than by handing out trial credits. What it does not publish is the rate limit that applies to them, so the list below separates what Z.AI states from what only outside trackers report.

• GLM-4.7-Flash, GLM-4.5-Flash, and GLM-4.6V-Flash are each listed at $0 for input, output, and cached input on the Z.AI pricing page as of 2026-10-07.
• GLM-5.3-Flash costs $0.15 per 1M input and $0.50 per 1M output. It is the cheapest current-generation model, not a free one.
• The free models are older generations. If you want GLM-5.3 quality at no marginal cost, the only route is self-hosting the open weights, and running GLM locally covers the hardware that takes.
• Z.AI's public pricing and docs pages do not publish a numeric rate limit for free-tier access, whether requests per minute, tokens per minute, or a daily cap. The page states only that standard API rate limits and Terms of Service apply, with no number attached. Check your own quota in the Z.AI console.
• Third-party API trackers report free-tier throughput of roughly 1 request per second, about 60 a minute, with a daily cap near 1,000 requests. That figure comes from outside observation rather than Z.AI's documentation, so treat it as directional and not contractual.
• Z.AI does not commit to keeping these models free. They are priced at zero today, and that can change without notice.
• For the paid models, Z.AI's docs mention a limited-time free benefit on cached-input storage. Storing cached prefixes is free during that promotional window, but the cached-input read price above still applies.
• There is no bundled free-credit grant on the public pricing page. If you need trial credits, higher free-tier throughput, or a written rate-limit commitment for a proof of concept, ask Z.AI directly through the subscribe page.
• For how to reach these models rather than what they cost, the Z.AI API guide covers keys, base URLs and endpoints, and our Z.AI review covers the data-retention terms a buyer has to ask for in writing. If you are looking for a phone or desktop client rather than an API, which Z.AI apps exist has the dated answer.

Enterprise / Custom Pricing

Z.AI does not publish enterprise-tier pricing. Volume discounts, dedicated deployments, private endpoints, data-residency options, and SLA-backed availability are all negotiated case-by-case.

• Volume discounts: not publicly listed — contact Z.AI sales for token-volume commitments.
• Dedicated / private deployment: not publicly listed — enterprise-only. Buyers focused on data isolation often self-host the open-weight GLM releases instead (see the self-hosted section below).
• Data-processing agreements and regional hosting: needed for GDPR-scope data. Not documented on the public pricing page — request through sales.
• Support SLAs: not publicly listed. Standard API access has documented rate limits but no uptime SLA on the public page.

How Z.AI Pricing Compares

Z.AI charges less for GLM-5.3 than Anthropic charges for Claude Sonnet 5.5 and Claude Opus 5.5, but more than DeepSeek charges for DeepSeek-V4-Pro. The nearest pairs are flagship GLM-5.3 against Claude Sonnet 5.5 and DeepSeek-V4-Pro, and lightweight GLM-5.3-Flash against Claude Haiku 4.5 and DeepSeek-V4.1-Flash. All figures reflect published rates in USD per 1 million tokens, verified against each provider's documentation on 2026-10-07.

For reasoning and coding tasks, GLM-5.3 is cheaper than Anthropic. At $1.40 for input and $4.40 for output, GLM-5.3 is 30% lower on input and 56% lower on output than Claude Sonnet 5.5 ($2 input, $10 output). Against Claude Opus 5.5 ($4 input, $20 output), GLM-5.3 costs 65% less on input and 78% less on output, while Z.AI bills cached input at $0.26.

Anthropic also lists Claude Fable 5.1 at $10 input and $50 output. For teams mapping prompt caching strategies or enterprise tiers across Anthropic models, see our Claude API pricing guide.

The cost balance flips when you compare GLM-5.3 to DeepSeek. GLM-5.3 costs more than DeepSeek-V4-Pro during peak hours ($1.40 versus $1.32 input, and $4.40 versus $3.96 output). During off-peak windows, DeepSeek-V4-Pro bills at half price ($0.66 input, $1.98 output), which makes GLM-5.3 roughly double the price of DeepSeek-V4-Pro.

DeepSeek schedules peak hours from 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, excluding Chinese public holidays. All remaining weekday hours and the entire weekend run at off-peak rates. You can examine DeepSeek's rate card in our DeepSeek pricing guide or our GLM vs DeepSeek comparison.

In lightweight models, GLM-5.3-Flash undercuts Claude Haiku 4.5 by wide margins. GLM-5.3-Flash costs $0.15 input, $0.50 output, and $0.03 for cached input, whereas Claude Haiku 4.5 bills $1 input and $5 output. That puts GLM-5.3-Flash at ten times lower cost on output tokens than Anthropic's entry model.

Against DeepSeek-V4.1-Flash (routed under the model name deepseek-flash), GLM-5.3-Flash holds a pricing edge. Its $0.15 input rate matches DeepSeek's off-peak input price and is half of DeepSeek's $0.30 peak input rate. On output, GLM-5.3-Flash at $0.50 sits below both DeepSeek's $0.60 off-peak rate and its $1.20 peak rate.

Z.AI also lists GLM-5.3-FlashX at $0.37 input, $1.25 output, and $0.075 cached input, alongside GLM-4.5-Air at $0.20 input and $1.10 output. For zero-cost testing, Z.AI provides GLM-4.7-Flash and GLM-4.5-Flash for free at $0 per token. For development teams evaluating subscription plans, see our GLM coding plan guide.

OpenAI changes its rates often, so check the OpenAI pricing page for current figures before you plan a production budget.

• Flagship tier, per 1M input / output tokens: GLM-5.3 $1.40 / $4.40, Claude Sonnet 5.5 $2 / $10, Claude Opus 5.5 $4 / $20, DeepSeek-V4-Pro $0.66 / $1.98 off-peak and $1.32 / $3.96 peak.
• Light tier, per 1M input / output tokens: GLM-5.3-Flash $0.15 / $0.50, Claude Haiku 4.5 $1 / $5, DeepSeek-V4.1-Flash $0.15 / $0.60 off-peak and $0.30 / $1.20 peak.

Self-Hosted Cost (Open Weights)

Several GLM releases (including GLM-4.5, GLM-4.5-Air, and GLM-5.2) ship as open weights under permissive licenses. Self-hosting means no per-token API cost, but you pay for GPUs, engineers, and operational overhead.

• Per-token cost: $0. There is no license fee for the open-weight GLM releases.
• Hardware class: flagship GLM models require a multi-GPU node — typically 8x H100/H200-class accelerators for full-precision serving. Quantized deployments can run on smaller footprints with quality trade-offs.
• GPU capital or rental: exact list prices are not published across cloud vendors — check the live pricing on your target cloud (AWS, GCP, Azure, Lambda, CoreWeave) before modeling. A single H100 rental typically runs several dollars per hour on-demand at public cloud vendors.
• Operational cost: engineering time to stand up vLLM/SGLang, monitoring, autoscaling, model updates, and security patching. These are the costs that surprise most teams.
• Break-even: self-hosting only wins on cost when steady-state token volume is very high, or when the driver is data isolation rather than pure economics.

Total Cost of Ownership

The sticker price on a pricing page is only one line of the real cost of running GLM in production. A realistic TCO model has to include the operating cost around the token bill.

• Token spend: per-1M rates above, multiplied by realistic input and output volumes for your workload — not the marketing example.
• Prompt caching: Z.AI's cached-input rate is roughly 5x cheaper than standard input for GLM-4.6. Structured prompts that reuse a long system prefix can meaningfully lower the effective per-request cost.
• Guardrails and moderation: teams routing production traffic to any Chinese-lab model typically add a moderation layer, log review, and prompt-filtering — this is engineering time and, often, a second API bill.
• Data-processing and compliance: GDPR-scope traffic requires a DPA or self-hosted deployment. Compliance review is a real line item, not a rounding error.
• Fallback provider: production stacks usually keep a second provider on standby for outages or rate-limit events. Budget for the second provider's minimums.
• Engineering time to swap: switching from Claude or GPT to GLM is not free — evaluation, prompt-tuning, and eval-set rebuilds all cost engineering hours.

When Z.AI's Pricing Wins

Z.AI's pricing is genuinely competitive, but not for every workload. The scenarios below are where GLM's per-token math translates into real savings.

• High-volume batch coding: overnight refactors, vulnerability scans, or agentic code review where per-token cost dominates and latency is not the constraint.
• Long-context document processing: legal review, contract analysis, or research summarization where GLM's long-context tiers avoid chunking overhead.
• Individual developers on the Coding Plan: at $18/month for Lite, the effective per-prompt cost is well below equivalent Claude Sonnet or GPT-5 usage inside coding IDEs.
• Free-tier experimentation: GLM-4.5-Flash at $0 is a real free model for prototyping, evaluation harnesses, and internal tooling.
• Cost-driven agent workloads: multi-step agents that burn tokens quickly benefit most from the lower per-1M rates.
• Scenarios where Z.AI does NOT win: real-time customer-facing chat where latency and moderation matter more than token price; regulated workloads that require a US-based provider with an existing BAA/DPA; teams already deeply invested in Claude or OpenAI tooling where switching cost exceeds the savings.

What you need to run GLM-5.2 yourself

GLM-5.2 is a frontier-scale Mixture-of-Experts model, so "running it yourself" is a real infrastructure decision — not something a single laptop or gaming GPU can do. Match the path below to how seriously you need to self-host. For most teams the API or rented GPUs are the right answer; buying hardware only pays off at steady, high volume or when your data can never leave your walls.

Once GLM-5.2 is running, the fastest way to put it to work day to day is inside Cursor — point it at the model through OpenRouter as a custom model. And if you would rather run a model on one affordable box, see Best mini PCs for local AI and Local AI hardware calculator.

NVIDIA RTX 6000 Ada (48GB)

Power users and small clusters that want cards they own

View on Amazon →