GLM Coding Plan: Prices, Credits and Claude Code Setup

The GLM Coding Plan is Z.ai's subscription for supported coding tools. The current rules use token-weighted credits, with separate five-hour and weekly allowances; the old fixed-prompt comparison is no longer a useful way to price it. This guide explains the allowance, how to avoid using the wrong endpoint, and what remains unverified about checkout prices. Official documentation was checked on October 4, 2026.

October 4 update: replaced the old prompt-count allowance and September promotion with the current credit rules. Removed unverified Pro/Max prices and automatic claims that the subscription is cheaper. This is a documentation and calculation update, not a new performance test.

What the subscription covers

The plan pays for GLM usage inside supported coding tools, including Claude Code, Cline and OpenCode. It is not an unrestricted API token budget for your website, scripts or gateway. Z.ai's plan FAQ restricts the benefit to supported tools and says those tools share the subscription quota.

Before troubleshooting a balance error, establish which product you are calling. For Claude Code, the documented plan base URL is https://api.z.ai/api/anthropic. For other supported tools, the FAQ lists https://api.z.ai/api/coding/paas/v4. These are not interchangeable with a general-purpose API route.

GLM Coding Plan prices and credit allowances

Z.ai's current usage instructions publish the following allowances. The overview advertises a starting price of $18/month, but we could not verify the complete tier-by-tier checkout on z.ai/subscribe. That starting figure is a vendor statement, not a verified price for your billing cycle or region.

Credits are not messages. Consumption depends on the model, uncached and cached input, output, and eligible off-peak discounts. The five-hour allowance refreshes dynamically after consumption; the weekly allowance resets on a seven-day subscription cycle. Check the subscription usage page for your remaining quota.

The old article described a promotion ending in September. We do not carry that forward as an October offer. The documentation now describes a 50% off-peak model-credit rate, but the retrieved rules do not specify an hour range. Check the live plan conditions rather than reusing the old peak-hour schedule.

Calculate credits for your own coding request

The published formula weights uncached input, cached input and output separately. For GLM-5.3, their respective multipliers are 6.9, 1.7 and 24; for GLM-5.3-Flash they are 2.3, 0.56 and 8. The weighted sum is divided by 10,000. These are quota multipliers, not dollar prices.

credits = (uncached_input * input_multiplier
         + cached_input * cache_multiplier
         + output * output_multiplier) / 10000

Worked example, not a measured session: suppose a turn uses 10,000 uncached input tokens, 20,000 cached input tokens and 1,000 output tokens, with no MCP calls. GLM-5.3 consumes (69,000 + 34,000 + 24,000) / 10,000 = 12.7 credits at the standard rate, or 6.35 at the documented off-peak rate. The same assumed token profile on Flash consumes 4.22 standard credits.

Lite's 2,000-credit five-hour allowance would cover about 157 such GLM-5.3 turns at the standard rate, assuming no other usage. That is scenario arithmetic, not a promised request cap: a larger repository context, more output, different cache hits or MCP usage changes the answer. Use your actual usage records before choosing a tier.

Claude Code setup: check the route before the balance

Follow Z.ai's Claude Code integration guide for your installed version. A shell configuration using the plan endpoint looks like this:

export ANTHROPIC_BASE_URL="https://api.z.ai/api/anthropic"
export ANTHROPIC_AUTH_TOKEN="YOUR_Z_AI_PLAN_KEY"
export ANTHROPIC_MODEL="glm-5.3"

claude

This illustrates the documented route; we did not run a new authenticated Claude Code session for this update. Keep the key out of screenshots, source control and shared logs. If an existing settings file overrides these values, inspect the active configuration before changing or deleting it.

For an 1113 Insufficient Balance message or unexpected metered deductions, check three things: the tool is supported, the base URL is the coding route, and the requested model is supported by the plan. A subscription receipt alone does not establish that a request used the subscription.

GLM-5.2 requests no longer prove GLM-5.2 was served

The plan overview currently names GLM-5.3 and GLM-5.3-Flash. It says GLM-5.2/5.1 requests are automatically routed to GLM-5.3, and GLM-4.7 requests to GLM-5.3-Flash. Use an explicit supported ID for new setup and retain the returned model/provider information when available.

Historical tests remain tests of the endpoint and version recorded on their run date. They do not become benchmarks of today's routed model simply because an older ID still accepts requests. Our GLM-5.3 review and measurement methodology keep that distinction separate from subscription benefits.

GLM API pricing is separate from plan credits

For application traffic, Z.ai publishes metered API rates. Selected text-model rates below are USD per million tokens, checked October 4. They are Z.ai direct API prices, not subscription prices or a gateway quote.

The plan's token multipliers are not these dollar rates. A third-party route can also have different prices, routing and billing terms; check its catalog rather than applying this table to another provider.

When is a subscription the better fit?

Choose based on the work you can actually complete inside the supported tools and quota windows. For steady coding, compare a verified monthly subscription price with your metered bill for the same workload. For an application or occasional scripting, the plan restrictions can make a metered API the relevant product regardless of the headline subscription price.

Keep a short log of uncached input, cached input, output, model and time of use. Price that log on both routes, include quota interruptions, and check renewal terms before buying. A large token allowance calculated at a high cache-hit rate is not a guarantee for a fresh repository.

For a second subscription option, see our Kimi Code plan guide. For provider-independent application access, DataLLM Lab's catalog is a separate metered product; this article does not claim its gateway can spend your GLM subscription credits.

FAQ