GLM-5.3 Coding Plan Provider for Claude Code

GLM-5.3 is a reasoning model available through z.ai’s Coding Plan. Its sonnet-tier companion GLM-5.3-Flash is Z.ai’s first native multimodal model in the GLM-5 series. Both sit behind a native Anthropic-compatible endpoint, so Claude Code talks to them without an adapter or a proxy.

Model specs

GLM-5.3-Flash is natively multimodal, reaches the full 1M-token context window through the [1m] suffix, and carries 3x the Coding Plan quota of GLM-5.3. That combination suits the day-to-day sonnet workload, since the same work drains far less of your balance.

The older glm-5.2 and glm-5.1 ids are no longer served on the Coding Plan endpoint. Requests for them are automatically routed to glm-5.3, so existing configs keep working, but the ids are effectively deprecated.

Thinking is always on: GLM-5.3 rejects requests that disable it, and supports reasoning_effort values of low, high, and max. The template sets REASONING_EFFORT to high and allocates 8,000 thinking tokens, which is enough for most code tasks without consuming much of the context window.

Context windows differ across tiers, so the template sets them per model instead of using one global override. The [1m] suffix on GLM-5.3 and GLM-5.3-Flash tells Claude Code their real 1M window in the opus and sonnet slots, and Claude Code’s default 200K assumption for unrecognized models already matches GLM-5-Turbo. Auto-compaction fires at the right point on every tier, so you do not need a CLAUDE_CODE_AUTO_COMPACT_WINDOW override.

Setup

1. Get a Coding Plan subscription at z.ai
2. Copy your API key from the dashboard
3. Run the setup command:

claude-multi add glm

4. Paste your API key when prompted

The template sets the base URL, model mappings, context limits, and thinking parameters.

When to pick GLM

Pick GLM when you want a fixed monthly cost instead of per-token billing. GLM-5-Turbo takes the light work (quick edits, shell commands, subagent calls), GLM-5.3 takes heavier refactoring and long agentic runs, and GLM-5.3-Flash covers day-to-day coding while draining the balance a third as fast as GLM-5.3. All of it runs inside one plan.

If your workload is bursty and you prefer paying only for what you use, look at the DeepSeek or MiMo pay-per-token templates instead.

Pricing details

GLM reaches Claude Code through Z.ai’s Coding Plan, a monthly subscription. The Anthropic-compatible endpoint is gated to the plan; there is no pay-per-token Anthropic URL for GLM, and GLM-5.3 is Coding Plan-only at launch with general pay-as-you-go API access coming soon.

Usage is metered in credits: (input x input_mult + cached_input x cached_mult + output x output_mult) / 10,000, with multipliers per model:

GLM-5.3-Flash is available on the Coding Plan and bills through the same points system rather than its own multiplier row. It carries 3x the GLM-5.3 quota, so sonnet-tier day-to-day usage drains the balance more slowly. Off-peak calls cost 50% of the standard rate, same as every other model on the plan.

Off-peak requests count at 50% of the standard credit rate. Peak is only Monday-Friday 14:00-18:00 UTC+8, so nights, weekends, and weekday mornings all bill at the off-peak rate.

Plans start from $18 USD/month for Lite, with 20% off quarterly and 30% off yearly billing. This points system replaced the older peak/off-peak multiplier plan, which was discontinued for new users on 2026-07-30. GLM-5-Turbo is also on the standard pay-as-you-go API, but that route isn’t Anthropic-compatible, so claude-multi doesn’t use it. For benchmarks and a full breakdown of the plan, see the GLM-5.3 announcement post. The GLM-5.2 post covers how the plan looked at launch, and z.ai has current Pro and Max pricing.

Related providers

• DeepSeek - pay-per-token, 1M context
• MiniMax - 1M context window, subscription
• Qwen - Alibaba’s coder models, pay-per-token