GLM 5.3 Coding Plan

TL;DR — GLM 5.3 is Z.ai's new flagship model, built on the GLM 5.2 base with heavy long-horizon post-training. It is already live for every GLM Coding Plan subscriber (Lite, Pro, and Max tiers), and Z.ai reports a 50% coding performance gain over GLM 5.2 on its in-house Z.ai Code Bench. The model API and pricing are still unpublished — for now, the GLM Coding Plan is the only way to use GLM 5.3 in coding tools like Claude Code, Cline, and OpenCode.

What Is the GLM 5.3 Coding Plan?

The GLM Coding Plan is a subscription package from Z.ai (Zhipu AI) designed specifically for AI-assisted coding. It gives you access to GLM models inside coding agents and IDEs through Anthropic- or OpenAI-compatible endpoints, instead of buying tokens per request.

GLM 5.3 is now available to all GLM Coding Plan users — the Lite, Pro, and Max tiers alike. According to Z.ai's official model documentation, the plan supports GLM-5.3, GLM-5-Turbo, and GLM-4.7, and requests for older models (GLM-5.2 and GLM-5.1) are automatically routed to GLM-5.3. That makes the coding plan the fastest way to try the new flagship today, ahead of the public model API.

GLM 5.3 vs GLM 5.2: What Actually Changed

GLM 5.3 keeps the same base model as GLM 5.2 — every improvement comes from post-training, not a new architecture. Z.ai reports the largest jumps on agentic and terminal benchmarks:

Z.ai says GLM 5.3 reaches state-of-the-art results among open-source models on Terminal-Bench 3.0 and Agents' Last Exam (CLI). On its internal Z.ai Code Bench, at "Max" reasoning effort GLM 5.3 scores 34.5% using roughly 75K output tokens per task, compared with GLM 5.2's 23.4% at around 96K tokens — better results at lower token cost. At "High" effort it reaches 31.4% at ~50K output tokens, ahead of Claude Opus 4.8's 29.5% at 120K.

Read the scores with a caveat: these are vendor-reported numbers for a specific harness and date. A benchmark score only tells you how a model performs on that benchmark — reproduce a task similar to your own workflow before making a team decision.

The Model Specs That Matter for Coding

• Context window: 1M tokens, enough for whole-repository work in one pass
• Maximum output: 128K tokens per response
• Modality: text-in, text-out
• Reasoning: always enabled, with three effort levels — low, high, max (you can no longer switch thinking off; the docs recommend max for complex coding)

GLM 5.3 Coding Plan Tiers and Credits

The plan uses a points-based quota instead of per-token billing. Each tier has a rolling 5-hour credit allowance and a weekly allowance:

Credit consumption is calculated from input, cached input, and output tokens weighted by model multipliers. GLM-5.3 uses an input multiplier of 6.9, a cached input multiplier of 1.7, and an output multiplier of 24 — the highest of the plan models, reflecting its deep-reasoning profile. Z.ai estimates that on GLM-5.3 with a typical 90.9% cache hit rate, Lite works out to roughly 43–87M tokens per week, Pro 263–526M, and Max 613–1,226M.

Two details worth knowing before you subscribe:

• Off-peak pricing: model usage is charged at 50% of the standard credit rate during off-peak hours — weekends all day, plus Monday–Friday outside 14:00–18:00 Singapore time (UTC+8).
• Entry price: the plan starts at $18/month (official documentation lists Lite as "starting at just 18 USD per month"); Pro and Max are aimed at high-frequency, complex projects.

How to Switch Your Coding Tool to GLM 5.3

If you already have a coding plan subscription, switching is a settings change in your agent — no new model ID required.

In Claude Code, add or update these environment variables in ~/.claude/settings.json (or the equivalent path for your OS):

{
  "env": {
    "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "1000000",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-4.7",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.3[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.3[1m]"
  }
}

The [1m] suffix enables the 1M-token context; pair it with CLAUDE_CODE_AUTO_COMPACT_WINDOW set to 1000000. In a session, type /status to confirm the model switched to glm-5.3, and use /effort to change thinking intensity. max is the default and the level Z.ai recommends for coding tasks.

In Cline (and most OpenAI-compatible tools), configure:

• API Provider: OpenAI Compatible
• Base URL: https://api.z.ai/api/coding/paas/v4
• API Key: your Z.AI API key
• Model: glm-5.3
• Context Window Size: 1000000 (disable image support)

The plan supports the same endpoints across tools: Anthropic-compatible tools (Claude Code, Goose) use https://api.z.ai/api/anthropic, Codex uses https://api.z.ai/api/v1, and OpenAI-compatible tools use https://api.z.ai/api/coding/paas/v4. Officially supported tools include Claude Code, Roo Code, Kilo Code, Cline, OpenCode, OpenClaw, Crush, Goose, and Cursor.

Who Should Buy the GLM 5.3 Coding Plan

Pick a tier based on how you actually work:

• Lite — solo developers doing daily coding tasks with light context; the entry price keeps the barrier low.
• Pro — teams working on whole codebases and long refactors where 1M-token context and 128K outputs matter.
• Max — high-frequency agentic work: long-horizon agents, automated task handling, and heavy MCP tool use.

One rule of thumb: if your work is mostly short snippets and quick questions, the credits go further on GLM-5-Turbo or GLM-4.7. Save GLM 5.3's higher multipliers for tasks that actually need deep reasoning — large refactors, cross-file debugging, and long-running agent sessions.

Frequently Asked Questions

Is GLM 5.3 available on the coding plan? Yes. Z.ai states GLM-5.3 is live for all GLM Coding Plan users (Lite, Pro, and Max), and requests for GLM-5.2/GLM-5.1 are auto-routed to GLM 5.3.

When is the GLM 5.3 API releasing? Not published. Z.ai says the model API is coming soon; no firm date, pricing, or official model-ID limits have been announced.

Is GLM 5.3 open-source? Weights and license are unconfirmed. Z.ai open-sourced GLM 5.2 under the MIT license on Hugging Face, so a release is widely expected — but treat it as unconfirmed until announced.

Does GLM 5.3 cost more on the plan? It uses higher credit multipliers than GLM-5-Turbo and GLM-4.7, but there is no extra subscription fee. Off-peak usage is charged at half credits.

Can I try GLM 5.3 without a coding plan? Yes. Third-party access like glm5.app lets you chat with GLM 5.3 in your browser with free chat credits, and track API, pricing, and open-source news in one place.

Try GLM 5.3 Today

The GLM 5.3 Coding Plan is the fastest official route to the new flagship in your daily tools. If you want a no-commitment look first, open GLM 5.3 chat on glm5.app, give it a realistic task from your own repo — a refactor plan with rollback steps, say — and judge the 1M-token context for yourself before you subscribe.

Start a free GLM 5.3 chat at glm5.app/glm-5-3, and subscribe to the official GLM Coding Plan if the results hold up in your workflow.

Sources

• GLM-5.3 model documentation (docs.z.ai) — Specs, benchmark figures, reasoning effort levels, and coding-plan availability. Vendor-reported numbers as of August 2026.
• GLM Coding Plan overview (docs.z.ai) — Plan tiers, credit allowances, multipliers, off-peak pricing, and $18/month entry price.
• How to Switch Models (docs.z.ai) — Claude Code, Cline, and endpoint configuration for GLM-5.3.
• GLM Coding Plan Quick Start (docs.z.ai) — Supported tools and setup steps.
• GLM Coding Plan subscription (z.ai) — Official plan tiers and sign-up.