Zhipu AI (Z.ai)
GLM 5.3
Zhipu's open-weight coding flagship — the GLM 5.2 base, post-trained into the strongest open-weights coding model they have published.
Maker
Zhipu AI (Z.ai)
What is GLM 5.3?
GLM-5.3 is Z.ai's current flagship, released 14 August 2026. It keeps GLM-5.2's 743B-parameter sparse MoE base (around 40B active per token) and 1M-token context window unchanged — every gain comes from an extended post-training pass, which Zhipu says lifts coding capability around 50%, with Terminal-Bench 3.0 moving from 4.6 to 28.3. It is priced identically to 5.2 on the z.ai API at $1.40 in / $4.40 out per million tokens, so the upgrade is free per token. One behavioural change matters to anyone calling it: reasoning is always on and cannot be disabled.
Strengths
- ✓Same price per token as GLM-5.2, with a large claimed coding gain from post-training alone.
- ✓Open weights (MIT) on a 1M-token-context MoE base — self-hostable and fine-tuneable.
- ✓Strongest open-weights agentic-coding scores Zhipu has published (DeepSWE 66.9%, Terminal-Bench 3.0 28.3).
Weaknesses
- ✗Thinking cannot be disabled — every call pays reasoning tokens and latency runs to minutes.
- ✗Post-training only: the 743B serving cost and Chinese-first tooling of the 5.2 base are unchanged.
- ✗Launch benchmarks are vendor-reported; third-party replication was still pending.
Top images made with GLM 5.3
Real generations from the community, most-liked first.
Prompts that work with GLM 5.3
Popular templates you can run with GLM 5.3 in one click.
Want to use GLM 5.3 on your own photos?
We host GLM 5.3 alongside every other major AI model. Free to try.
Open the GLM 5.3 prompt library →