How to Use GLM-5.3 in Claude Code (Official Z.AI Setup, 2026)
Run Claude Code on GLM-5.3 and GLM-5.3 Flash with Z.AI's official Anthropic-compatible endpoint. Exact settings.json, which API key to use (pay-as-you-go vs GLM Coding Plan), and worked cost math against Claude Opus 5.5.
Z.AI (Zhipu) documents how to run Claude Code on GLM-5.3 through its Anthropic-compatible endpoint. You can pay per token with a standard API key, or use a GLM Coding Plan subscription key. The settings are the same; the key decides how you’re billed.
Everything below follows Z.AI’s Claude Code guide, model switching guide and API key management guide, checked on 2026-11-16. Prices come from Z.AI’s pricing page and Anthropic’s pricing page as recorded in DevTk.AI’s model data.
Who supports this: Z.AI documents and supports this setup. Anthropic’s Claude Code docs say Anthropic doesn’t support routing Claude Code to non-Claude models.
Quick answer
| Setting | Value |
|---|---|
| Endpoint (Z.AI, international) | https://api.z.ai/api/anthropic |
| Endpoint (BigModel, mainland China) | https://open.bigmodel.cn/api/anthropic, per Zhipu’s Chinese guide |
| Opus and Sonnet tiers | glm-5.3[1m] |
| Haiku tier | glm-5.3-flash[1m] |
| Where to put it | the env block of ~/.claude/settings.json |
| Billing | Decided by the key type (see below) |
Step 1: pick the right API key
Since 2026-10-16, Z.AI manages keys for the standard API and for the GLM Coding Plan separately, and for keys created from that date the key type decides billing, whatever endpoint you call:
| You want | Key type | Billing |
|---|---|---|
| Pay per token | Standard API key | Your account balance or usage bundles; doesn’t use Coding Plan quota |
| Use your GLM Coding Plan | Personal Coding Plan key | Your plan’s 5-hour and weekly quota |
| Use a team plan | Team Plan key | Team plan and seat rules |
Z.AI’s example: a standard key used in a coding tool bills pay-as-you-go even if you also have a Coding Plan. Keys created before 2026-10-16 were classified automatically; check the label in the Z.AI console if you’re unsure.
Step 2: install Claude Code
npm install -g @anthropic-ai/claude-code
cd your-project
claude
Z.AI lists Node.js 18 or newer, and Git for Windows on Windows.
Step 3: configure settings.json
Z.AI’s manual configuration (add or update these keys in ~/.claude/settings.json; don’t replace other fields):
{
"env": {
"ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
"ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5.3-flash[1m]",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.3[1m]",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.3[1m]",
"CLAUDE_CODE_AUTO_COMPACT_WINDOW": "1000000",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1,
"DISABLE_TELEMETRY": "1",
"CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT": "0",
"CLAUDE_CODE_MAX_CONTEXT_TOKENS": "1000000",
"API_TIMEOUT_MS": "3000000"
}
}
Open a new terminal, run claude in your project, and answer Yes if Claude Code asks “Do you want to use this API key”. The file holds your key in plain text; don’t commit it.
To run everything on the cheaper Flash model, Z.AI’s switching guide sets all three tiers to glm-5.3-flash[1m].
Z.AI also offers a setup helper (npx @z_ai/coding-helper) and a macOS/Linux script that write the same file. The manual version above shows exactly what changes.
What changes on GLM-5.3
| Area | What happens |
|---|---|
| Your Claude plan | With a Z.AI key in ANTHROPIC_AUTH_TOKEN, Claude Pro/Max/Ultra limits aren’t used. |
| Reasoning | GLM-5.3 always reasons, with low, high and max effort levels. |
| Telemetry | Z.AI’s config turns off Claude Code’s non-essential traffic and telemetry. |
| Coding Plan tools | Z.AI says the GLM Coding Plan may only be used in its officially supported tools. Claude Code is one of them. |
Cost: pay-as-you-go
Claude Code defaults to Claude Opus 5.5 on paid plans and the API, per Anthropic’s model configuration docs.
Assumed workload: 100M input tokens and 2M output tokens a month, 90% of input as cache hits (90M cached, 10M uncached), all on the Opus/Sonnet tier. Anthropic’s cache-write premium is left out.
| Model | Uncached input | Cached input | Output | Monthly total |
|---|---|---|---|---|
| Claude Opus 5.5 | 10M × $4.00 = $40.00 | 90M × $0.20 = $18.00 | 2M × $20.00 = $40.00 | $98.00 |
| Claude Sonnet 5.5 | 10M × $2.00 = $20.00 | 90M × $0.10 = $9.00 | 2M × $10.00 = $20.00 | $49.00 |
| GLM-5.3 | 10M × $1.40 = $14.00 | 90M × $0.26 = $23.40 | 2M × $4.40 = $8.80 | $46.20 |
| GLM-5.3 Flash | 10M × $0.15 = $1.50 | 90M × $0.03 = $2.70 | 2M × $0.50 = $1.00 | $5.20 |
At pay-as-you-go rates GLM-5.3 costs a bit under half of Opus 5.5 on this workload, but about the same as Sonnet 5.5, because its cached-input price ($0.26/M) is higher than Sonnet’s ($0.10/M) and cache reads dominate agent traffic. GLM-5.3 Flash is the cheap option at about 5% of Opus.
Cost: GLM Coding Plan credits
Z.AI publishes how the GLM Coding Plan meters usage, which lets you estimate whether a workload fits:
- Credits used = (input tokens × input multiplier + cached tokens × cached multiplier + output tokens × output multiplier) / 10,000.
- GLM-5.3 multipliers: 6.9 input, 1.7 cached input, 24 output. GLM-5.3 Flash: 2.3, 0.56, 8.
- Off-peak usage costs 50% of the standard credit rate. Peak is Monday to Friday, 14:00-18:00 Singapore time (UTC+8).
| Plan | 5-hour credits | Weekly credits |
|---|---|---|
| Lite | 2,000 | 10,000 |
| Pro | 12,000 | 60,000 |
| Max | 28,000 | 140,000 |
The same workload on GLM-5.3, all at the peak rate:
- Uncached input: 10,000,000 × 6.9 / 10,000 = 6,900 credits
- Cached input: 90,000,000 × 1.7 / 10,000 = 15,300 credits
- Output: 2,000,000 × 24 / 10,000 = 4,800 credits
- Total: 27,000 credits a month, about 6,200 a week over 4.33 weeks.
That fits under Lite’s 10,000 weekly credits if spread evenly, and off-peak use would halve it. The 2,000-credit 5-hour cap on Lite is the tighter limit for bursty sessions. Z.AI’s docs say plans start at $18 a month but don’t list the Pro and Max prices; check the subscription page before you buy.
Troubleshooting
| Symptom | Fix |
|---|---|
| Settings don’t take effect | Close all Claude Code windows and open a new terminal; check the JSON is valid |
| Plan quota not used | You may be using a standard key; create a Coding Plan key under Personal Coding Plan |
| Unexpected pay-as-you-go charges | Same cause: the key type, not the endpoint, decides billing |
| Old model names | Requests for GLM-5.2/5.1 route to GLM-5.3 on the Coding Plan, but set glm-5.3 explicitly |
Related
Next step
Related Posts
How to Use Kimi K3 in Claude Code (Official Kimi Setup, 2026)
2026-11-16
Codex + DeepSeek V4: How to Use DeepSeek in Codex CLI (Official Setup, 2026)
2026-11-16
GLM-5.3 vs DeepSeek V4.1 vs Kimi K3/K3.1 Pro: API Price, Caching and Agent Routing
2026-08-17