How to Migrate from GLM 5.2 to GLM 5.3: Thinking Parameters & Breaking Changes

Quick answer: Migrating to GLM 5.3 requires exactly one breaking change: thinking.type: "disabled" is no longer supported — switch to "enabled" and set reasoning_effort (low/high/max, default max). Everything else — model ID, context, function calling, streaming — carries over. Do the thinking fix before updating the model ID, or requests fail.

TL;DR

The One Breaking Change

GLM 5.3 removes the ability to disable thinking. If your application sends:

{ "thinking": { "type": "disabled" } }

…the request fails once you switch to glm-5.3. The fix:

{
  "model": "glm-5.3",
  "thinking": { "type": "enabled" },
  "reasoning_effort": "max"
}

reasoning_effort values:

Z.AI recommends max for coding tasks. low exists for latency/cost-sensitive workloads — but note there is no "no thinking" mode anymore.

Migration Order (To Avoid Failed Calls)

1. Keep your current model ID (glm-5.2).
2. Change thinking.type to "enabled".
3. Set reasoning_effort: "low" as a safe starting point.
4. Update the model ID to glm-5.3.
5. Validate on your longest-running workloads, then raise effort to high/max as appropriate.

Z.AI's official migration note is the same: change it to enabled and set reasoning_effort to low before updating the model ID to glm-5.3. Otherwise, the request will fail.

What Else Changes (Minor)

• Tool calling now supports streaming parameter output — if you parse tool calls, verify your streaming handler tolerates parameter tokens arriving incrementally.
• Capability jump is the point: ~+50% on Z.AI's code bench, Terminal-Bench 3.0 4.6 → 28.3, AutomationBench 26.2 → 48.2. Your agent code doesn't need changes to benefit — the same prompts get better results.
• Model ID on OpenRouter: z-ai/glm-5.3 isn't listed yet (as of Aug 14, 2026) — the API itself is "coming soon." Coding Plan and ZCode users get 5.3 immediately.

What Stays the Same

• 1M-token context window, 128K max output — no prompt-packing changes needed.
• Function calling — same interface (plus the streaming-output upgrade).
• Streaming, structured output, temperature — unchanged.
• Base model behavior — same base as 5.2, so embedding/rag behavior is stable.

Coding Plan Migration (No Code Changes)

GLM Coding Plan subscribers were automatically upgraded to GLM 5.3 on launch day — nothing to configure. Inside ZCode / Claude Code / Cline / Kilo Code, the plan now routes to GLM 5.3. Two plan-side notes:

• The plan moved to a points-based quota (input/cached/output metered separately).
• Off-peak hours (outside 14:00–18:00 UTC+8 weekdays) consume 50% of standard points — schedule heavy runs accordingly.

Rollback Plan

If a workload misbehaves after cutover:

• API: revert the model ID to glm-5.2 — your old request format (including thinking.type: "disabled", if you kept it) works unchanged.
• Coding Plan: there's no per-request model toggle documented; contact Z.AI support for a rollback window if needed.

Checklist

• thinking.type set to enabled in all request templates
• reasoning_effort chosen per workload (max for coding)
• Streaming tool-call parser handles parameter streaming
• Staging run on glm-5.3 with low before full cutover
• Off-peak scheduling set for point-heavy jobs
• Rollback model ID documented in ops runbook

FAQ

Is GLM 5.3 backward compatible with GLM 5.2? Almost — one breaking change: thinking.type: "disabled" is removed. Fix that one field and the rest of the API carries over.

What does reasoning_effort do? Controls thinking depth: low (light), high (enhanced), max (deep, default). max is recommended for coding.

Can I still disable thinking in GLM 5.3? No — thinking is always enabled. low is the lightest setting.

Does GLM 5.3 change my context window? No — same 1M-token context and 128K max output as GLM 5.2.

Do I need to update my code for tool calling? Only if you parse tool calls from streams — GLM 5.3 adds streaming parameter output.

I'm on the Coding Plan — do I need to migrate? No — Coding Plan users were upgraded to GLM 5.3 automatically.

Sources

• Z.AI Docs: GLM-5.3
• Z.AI: GLM-5.3: Frontier Coding with Emergent Cyber Capabilities (August 14, 2026)

Last updated: August 14, 2026