ZCode vs Claude Code: cheap AI coding and your margin
Z.ai's GLM-5.2 builds software at roughly a fifth of Claude's price. Here is the real cost math, the hidden catch, and what to move to the cheap tool.
Some links below are partner links. We may earn a commission at no extra cost to you. How we make money.
The cost of having custom software built just fell by roughly 80% on paper. GLM-5.2, the model inside Z.ai's new free ZCode app, runs at about $4.40 per million tokens of output against Claude Opus 4.8's $25. It scores 62% to Claude's 69% on the standard test of fixing real bugs in real projects. For any operator paying to build landing pages, funnels, or internal tools, this is a cost line repricing in front of you.
The money angle
Two numbers set the frame. Usage price and plan price.
On usage, GLM-5.2 is $1.40 in and $4.40 out per million tokens (a token is roughly three-quarters of a word). Claude Opus 4.8 is $5 and $25. Claude's own new top model, Fable 5, is $10 and $50. So the spread from cheap to frontier is about seven times on output.
On plans, the gap is smaller than it looks. The ZCode app is free, and the Z.ai coding plan behind it starts near $18 a month. Claude Code rides on a $20 Pro plan, or a Max plan from $100 for heavy use. Entry to entry, that is close. The real savings show up at volume, where usage price does the work.
| Cost line | GLM-5.2 (ZCode) | Claude Opus 4.8 |
|---|---|---|
| Output, per million tokens | $4.40 | $25 |
| Entry plan, per month | free app + ~$18 | $20 |
| Heavy plan, per month | ~$160 | $100 to $200 |
One published estimate puts a workload costing about $1,000 a day in Claude Opus output near $176 a day on GLM-5.2. Same volume of output. A fifth of the bill.
Where it breaks
Cheaper per word is not cheaper per finished job. This is the line that costs people money.
The cheap model is weaker on long, multi-step work. On tests of sustained jobs it trails Claude by roughly half, so it retries more and burns more tokens to reach the same finished result. A "five times cheaper" price can quietly become "twice as many attempts."
Two more leaks. The Z.ai plan draws down faster during peak hours, so the $18 tier does less real work than the headline quota implies. And it is a Chinese cloud service, which raises data-residency questions the moment client information is involved. As one analyst put it, cheaper intelligence only helps if you can actually put it to work, and the setup around the model is the part most teams have not built.
Your move
What to watch
The cheap model already runs inside Claude Code, so you can keep the stronger tool and feed it the cheaper model for routine work, on one bill. Prices keep falling: Fable 5 reset the top of the market on June 9. For a numbers-driven operator, this is now a budgeting decision, not a capability cliff. See the pricing traps in other tools, the numbers worth watching, and our other cost breakdowns for where else the same math shows up.
Frequently asked questions
How much cheaper is GLM-5.2 than Claude for coding?
On raw usage, GLM-5.2 runs at $1.40 per million tokens in and $4.40 out, against Claude Opus 4.8 at $5 and $25. That is roughly a fifth of the output price. The ZCode app is free; Claude Code needs a $20 Pro plan or a Max plan from $100. Pricing checked July 2026.
Is cheaper per token actually cheaper per finished job?
Not always. The cheaper model is weaker on long, multi-step work and retries more on hard tasks, so it can burn more tokens to reach the same result. Measure cost per finished job on your own work, not the sticker price per million tokens.
What is the catch with the cheap coding plan?
Three things. Peak-hour usage on the Z.ai plan draws down your quota faster, so the entry tier does less than the number suggests. It is a Chinese cloud service, which raises data questions for client projects. And it trails Claude on sustained agent jobs, where the savings can evaporate.
Where does the cheap tool make sense for an operator?
High-volume, single-shot work: quick pages, small scripts, one-off fixes. Keep the premium tool for anything long-running, complex, or client-sensitive. You can also run the cheap model inside Claude Code, keeping one tool and one bill.
About The Margin
The brief on marketing that actually pays. Funnels, CRM, paid ads and lead gen for operators who care about ROI. How we work
The Margin · Daily Brief
Know what changed before it costs you
Each weekday: one shift in marketing, what it does to your numbers, and the move that protects them.
Short, priced, weekday mornings. Out in one click if it stops paying.
Keep reading
Related briefings
GoHighLevel vs HubSpot vs ClickFunnels for agencies
The agency CRM choice comes down to client volume and budget. GoHighLevel bundles everything; HubSpot scales reporting; ClickFunnels just does funnels.
Google reviews vanish, Hotel Ads shift: the July 4 brief
Google review bugs and Hotel Ads feed changes move local pipeline and paid travel spend. The July 4 brief tracks the budget risk.
Google tightens CPA targets, Bing backfills: the July 3 brief
Google Ads is changing budget-limited bidding while Bing backfills AI reports. The July 3 brief tracks CAC, attribution and brand visibility.