GLM-5.3 Coding Plan Provider for Claude Code

GLM-5.3 is Z.ai’s flagship coding model, sold through the GLM Coding Plan instead of a metered Anthropic API. Z.ai documents GLM-5.3 and GLM-5.3-Flash as the two models on the plan. Requests for GLM-5.2 and GLM-5.1 route to GLM-5.3, and GLM-4.7 requests route to GLM-5.3-Flash. The endpoint at api.z.ai/api/anthropic speaks Anthropic Messages directly, so Claude Code connects without an adapter.

Model specs

The template maps the Claude Code roles to two GLM models. Primary, opus, sonnet, and fable use GLM-5.3. Haiku uses GLM-5.3-Flash.

GLM-5.3 is text-only and always reasons. GLM-5.3-Flash is the first native multimodal model in the series, with text, image, video, and file input. It has 320B total parameters and 18B active, and it carries 3x the Coding Plan quota of GLM-5.3.

Both models have a 1M window. The [1m] suffix is a Claude Code context hint, not part of the model name sent to Z.ai. It tells Claude Code the real window. Without it, Claude Code assumes 200K for an unrecognized model and compacts early. The template also sets CLAUDE_CODE_AUTO_COMPACT_WINDOW to 1000000, so both slots compact at the same point.

Thinking cannot be disabled on GLM-5.3 or Flash. reasoning_effort accepts low, high, and max. Z.ai defaults to max, while the template sets high and 8,000 thinking tokens to keep latency down.

Template status: current

The 0.13.2 template uses the mapping from Z.ai’s manual Claude Code configuration: glm-5.3[1m] in primary, opus, sonnet, and fable, glm-5.3-flash[1m] in haiku, and the Coding Plan endpoint. It sets auto-compact to 1000000, CLAUDE_CODE_MAX_OUTPUT_TOKENS to 128000, effort to high, CLAUDE_CODE_ALWAYS_ENABLE_EFFORT to 1, MAX_THINKING_TOKENS to 8000, and disables nonessential traffic. No manual edits are needed after claude-multi add glm.

Z.ai’s current default configuration maps opus, sonnet, and haiku to glm-5.3-flash[1m]. The template keeps the manual split so the main slots use the flagship model. MAX_THINKING_TOKENS is a client-side hint; Z.ai derives thinking depth from the effort level.

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 writes the base URL, model mappings, context limits, and thinking parameters into ~/.claude-<name>/settings.json.

Pricing details

Usage is metered in credits: (input x input_mult + cached_input x cached_mult + output x output_mult) / 10,000. The multipliers differ by model:

The plan supports GLM-5.3 and GLM-5.3-Flash. Z.ai’s credit table lists multipliers for those two models.

At standard rates, GLM-5.3 consumes exactly 3x the credits of Flash: 6.9 vs 2.3 on input and 24 vs 8 on output. Cached input is the cheapest line on both models, which rewards long sessions that resend the same files.

Plans start at $18/month for Lite. Off-peak calls bill at 50% of the standard credit rate. Peak is Monday to Friday, 14:00-18:00 UTC+8, which is 06:00-10:00 UTC. That window lands in the European morning and late night on the US west coast, so check it before scheduling a long batch. Nights, weekends, and weekday mornings and evenings in Asia bill off-peak.

The GLM-5.3 announcement post has benchmarks and a plan breakdown. Z.ai’s Coding Plan docs list the live quotas and multipliers.

What to watch

GLM-5.3 launched as a Coding Plan model. Z.ai’s docs present the Anthropic endpoint as part of the plan, and the metered GLM API uses an OpenAI-compatible URL. There is no pay-per-token Anthropic route for GLM today, so this template requires a subscription. If Z.ai opens a metered Anthropic endpoint, claude-multi will need a second GLM template. Pointing the current template at the OpenAI URL will not work.

GLM-5.3-FlashX, the 200 tokens/s variant, is not on the Coding Plan yet. The plan carries GLM-5.3-Flash, not FlashX.

Related providers

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