GLM 5.3 Is Live — Direct Z.ai, Neck And Neck With Fable

August 19, 2026

Updated August 20, 2026: GLM 5.3 now routes directly through Z.ai's own API instead of OpenRouter. Same model, pricing, and pins — details below reflect the direct route.

GLM 5.3 replaces GLM 5.2 as Z.ai's coding and long-horizon agent flagship, with always-on reasoning, a 128K output cap, and Auto quality scored even with Claude Fable 5. Every call goes to Z.ai's own API.

What you can do

• Pick GLM 5.3 from the model selector — one entry, 1M-token context, tools, Swarm, and SOTA-tier Auto.
• Run always-on reasoning at max effort — Alfrada sends Z.ai's thinking.type: enabled and reasoning_effort: max on every call.
• Use the full 128K output window — up from GLM 5.2's 65K cap — for long refactors, multi-file plans, and extended agent traces.
• Keep Premium-band pricing (6×) — the same token band as GLM 5.2. Plan price, credit packs, and band multipliers are unchanged.

Where this shows up

You were pinning GLM 5.2 (direct Z.ai or OpenRouter) for project-scale engineering or long agent loops. The next message rolls forward to GLM 5.3 — no re-pin.

You pinned GLM 5.3 during the brief launch-day OpenRouter path. That pin rolls forward to the direct Z.ai route automatically.

You leave the picker on Auto. When the router escalates to SOTA, GLM 5.3 now sits in the same quality band as Claude Fable 5 and Claude Opus 5, so speed — not a quality handicap — decides the tie.

On the production harness it ranks #5 on Can it Run Alfrada? (8.55 composite) — a step up from GLM-5.2 at 7.95, behind Fable on calibrated rank.

Try it

• [Pin GLM 5.3] "Refactor this module end-to-end — read the whole repo context, plan, then implement."
• [Pin GLM 5.3] "Audit this codebase for security issues and produce a prioritized fix list."
• Leave the picker on Auto, then: "Research this market end to end and turn the findings into a board deck."

Heads up

• GLM 5.2 leaves the public picker. Sessions pinned to GLM 5.2, GLM 5.2 (via OpenRouter), or the launch-day GLM 5.3 (via OpenRouter) path roll forward to direct Z.ai GLM 5.3 automatically.
• Billing stays on Premium 6× — same token band as GLM 5.2. Your plan price, credit packs, and band multipliers are unchanged.
• Thinking cannot be disabled. Alfrada always sends max-effort reasoning.
• This is not the OpenRouter ZDR pin. Calls go to Z.ai's API (api.z.ai).