GLM-5.3 Pricing: What It Costs to Use Z.ai's New Coding Model

Z.ai launched GLM-5.3 on August 14, 2026, and if you are trying to price it out, the answer depends on how you plan to use it. The GLM Coding Plan is live and starts at $18 a month. The standard per-token API rate, the figure most cost calculators want, has not been published yet.

That split matters more than it sounds. If you code inside a tool like Claude Code, Cline, or Z.ai's own ZCode, you can subscribe to GLM-5.3 today at a fixed monthly price. If you were planning to call the model directly through the API and pay per token, you are waiting on Z.ai to publish that rate. For the full launch context, see our GLM-5.3 launch coverage. This guide covers what you can actually pay right now, how the Coding Plan credit system works, and what is still missing.

How much does GLM-5.3 cost?

GLM-5.3 costs $18 a month on the entry Coding Plan, rising to $80 for Pro and $168 for Max on monthly billing. Those are the only official GLM-5.3 prices Z.ai publishes today. There is no separate per-token rate for the model yet, so the Coding Plan is the practical way to pay for it.

The plan is a flat subscription rather than pure usage billing. You pick a tier, you get a weekly credit allowance, and you spend those credits as you code. This suits the audience Z.ai is targeting with GLM-5.3, which is developers running long coding sessions inside agent tools, not teams wiring the model into a product through raw API calls.

Z.ai offers 3 billing terms, and the discount grows the longer you commit:

• Monthly: full price, no discount
• Quarterly: 20% off
• Yearly: 30% off

Also read our GLM 5.2 benchmark guide to see how the model family performs before you commit to a plan.

GLM-5.3 Coding Plan pricing tiers

The 3 Coding Plan tiers differ mainly in how many credits you get each week, not in which model you can use. Every tier runs GLM-5.3. Lite is built for light iteration on small repositories, Pro for daily development on mid-sized codebases, and Max for heavy work on larger projects.

GLM-5.3 Coding Plan pricing, as of August 2026

The yearly column reflects the 30% annual discount billed upfront. A quarterly option sits between the two at 20% off. Pro adds a curated set of MCP tools and faster generation speeds on top of everything in Lite. Max adds dedicated resources during peak times and first access to new models.

One detail worth flagging: Pro and Max are defined by a multiple of Lite's usage rather than a flat credit number. Pro gives 6 times Lite's allowance and Max gives 14 times, so the effective weekly credits scale from Lite's 10,000 baseline. All 3 tiers support more than 20 agent tools, including ZCode, Claude Code, and others.

How the GLM-5.3 credit system works

GLM-5.3 bills through points, not a flat request count, so a single credit allowance stretches differently depending on what you do. The new Coding Plan charges input tokens, cached input tokens, and output tokens at separate rates, then converts them into credits. Output is by far the most expensive component.

Here is the mechanic in one line, taken from Z.ai's developer docs: your model credit usage equals input tokens times the input multiplier, plus cached input times its multiplier, plus output tokens times the output multiplier, all divided by 10,000.

The multipliers for GLM-5.3 are:

• Input: 6.9 per token unit
• Cached input: 1.7 per token unit
• Output: 24 per token unit

The gap between input and output is the number to internalize. Output tokens cost roughly 3.5 times what fresh input costs, and cached input is cheaper still, so reusing context is much cheaper than generating new text. For coding agents that read large files and return focused edits, that ratio works in your favour.

Peak and off-peak pricing

Z.ai charges half price outside a narrow weekday window, which is the single biggest lever on your effective cost. Any usage outside peak hours consumes 50% of the standard credit rate. Peak hours are Monday to Friday, 14:00 to 18:00 Singapore time (UTC+8). Everything else, including all of Saturday and Sunday, bills at the off-peak rate.

For a developer outside East Asian business hours, this is close to a permanent discount. If your working day rarely overlaps that 4-hour afternoon window in UTC+8, most of your usage lands in the half-price band by default.

What your credits actually buy

Credits are abstract, so Z.ai publishes estimated token allowances to make them concrete. Assuming all usage runs on GLM-5.3 with a 90.9% cache hit rate, which Z.ai calls the average for coding work, the estimated weekly token allowances are:

• Lite: 43 to 87 million tokens a week
• Pro: 263 to 526 million tokens a week
• Max: 613 to 1,226 million tokens a week

Each tier shows a range because of the peak and off-peak split. The high end assumes all usage falls in off-peak hours at half rate, and the low end assumes all of it lands during peak hours at full rate. Real usage sits somewhere between the two.

What about the GLM-5.3 API price?

The GLM-5.3 per-token API rate is not published yet, and you should not assume it matches GLM-5.2. This is the most important caveat in any GLM-5.3 pricing discussion right now. Z.ai's official API pricing table still lists GLM-5.2 as the newest entry, at $1.40 per 1M input tokens, $0.26 per 1M cached input tokens, and $4.40 per 1M output tokens. There is no GLM-5.3 row.

Z.ai has been explicit that API access is rolling out in stages behind a safety review, alongside the open weights. So the API exists as a route, the model ID is live in integration guides, but the standalone per-token price a cost model needs has not been posted. Treating GLM-5.2's numbers as a stand-in would be a guess, and Z.ai's own launch notes caution against it.

If per-token API pricing is a hard requirement for your budget, the honest answer is to wait for Z.ai to update the table. If you can work inside a supported coding tool in the meantime, the Coding Plan gives you GLM-5.3 access today at a known price. For how a fully-priced model breaks down by comparison, see our guide to Claude Opus 5 pricing.

When will GLM-5.3 be available to self-host?

GLM-5.3 open weights are promised roughly 2 weeks after launch, once Z.ai completes a safety evaluation. Until then, self-hosting is not possible. There is no downloadable checkpoint, no published license, and no official serving recipe for the exact model. Z.ai's launch page lists the weights as coming soon.

This is a real change from the GLM-5.2 cycle, where weights landed on Hugging Face under an MIT license within days and the community built cheap third-party access around them quickly. GLM-5.3 explicitly gates that step. Z.ai ties the delay to its cybersecurity focus for this release, framing the staged rollout as a safety measure rather than a scheduling accident.

If your plan depends on running the model on your own hardware, budget for that safety-review delay rather than a same-week drop, and watch Z.ai's Hugging Face org for the release.

Is GLM-5.3 worth the price?

GLM-5.3 is a clear upgrade over GLM-5.2 for coding and agent work, and at $18 a month the entry tier is priced to undercut most closed-model coding subscriptions. Whether it is the right model depends on what you are optimizing for, so it helps to read the benchmarks with a clear eye on where the numbers come from.

Z.ai reports that GLM-5.3 uses the same base model as GLM-5.2, with every gain coming from additional post-training. On its in-house Z.ai Code Bench, the company claims a 50% relative improvement over GLM-5.2 at high effort. That figure is vendor-reported on a private benchmark, so it cannot be independently reproduced. The direction is consistent across Z.ai's published table, but the exact margin is the company's own measurement.

GLM-5.3 vs peers on selected benchmarks, all figures vendor-reported by Z.ai unless noted

Two honest readings come out of that table. First, GLM-5.3 improves on GLM-5.2 across every row, which is the comparison Z.ai most wants you to make, and it genuinely holds. Second, GLM-5.3 does not lead the field. GPT-5.6 Sol posts higher numbers on the raw coding benchmarks, and GLM-5.3's clearest wins cluster around automation and defensive security rather than general coding.

A sourcing note matters here. Every score in Z.ai's launch chart is vendor-run on a public benchmark, with one exception: the GDPval-AA v2 Elo rating was produced by Artificial Analysis, an independent evaluator. Z.ai also ran its comparisons through the Claude Code harness at max reasoning effort, so these are not identical-harness races across every model. Read the table as directional, not as an audited leaderboard. For how the previous generation stacked up against a key rival, see our GLM 5.2 vs Kimi K3 comparison.

Where GLM-5.3 pricing leaves you today

GLM-5.3 pricing comes down to a single practical choice: if you code inside a supported tool, subscribe to the Coding Plan and you have GLM-5.3 at $18 a month today. If you need per-token API pricing or self-hosted weights, you are waiting a couple of weeks for Z.ai to publish the rest. For most developers testing a new coding model, the Coding Plan is the faster and cheaper path, especially with the off-peak discount doing quiet work in the background.

The bigger picture is that picking a model is only one decision in shipping an actual product, and often not the one that slows people down. If your goal is to turn an idea into a working, deployed app rather than to benchmark models against each other, the model is a means, not the project.

If you are weighing your options there, our roundup of the best AI coding tools is a useful starting point. That is where a platform like Emergent fits in.

Emergent lets you build and deploy full-stack apps through plain conversation, with access to leading models from OpenAI, Anthropic, and Google handled for you through a single Universal LLM Key, so you can focus on the product instead of wiring up APIs and comparing token rates. Skip the pricing math and ship your idea on a model that is ready today.

Start Building on Emergent.