Model cards

glm-5.3

Vendor zhipu Tier - Price per million tokens, in / out - Released 2026-08-14 Context 1,000,000 Max output 128,000 Vendor page

Last updated 2026-10-26 How this is made

Meant to excel at

• Z.ai reports Terminal-Bench 2.1 88.2 (level with Kimi K3 and GPT-5.6 Sol) and CyberGym 84.5, up from 77.2 on GLM-5.2 (vendor-relayed via DataNorth).

  Vendor claim, relayed datanorth.ai

• Arena WebDev 1621 and AA index 45, roughly level with other leading Chinese models.

  Independent evaluator arena.ai

Watch out for

• Token-hungry in a one-shot HTML animation test: about 47K to 54K output tokens per task including reasoning, the highest among four Chinese models tested.

  Practitioner report empiriolabs.ai

• Launch left API price unlisted and weights promised for later; verify availability and licence before planning around it.

  Vendor claim, relayed datanorth.ai

Benchmarks

Coding percentile 0.14

Agentic coding percentile 0.53

Overall percentile 0.35

Percentile is the share of tracked models this model beats on that benchmark, averaged across the group. Only benchmarks at least four tracked models share are scored.

Practitioner reports

• -

  On a second set of three one-shot HTML scenes it wrote the longest files (about 710 lines) and about 54,000 tokens per task.

  Compared with kimi-k3 | 2026-10-05

  Practitioner report empiriolabs.ai
• -

  At max reasoning effort GLM 5.3 spent about 47,000 output tokens per task on four single-file HTML animations, more than Kimi K3, Qwen3.8 Max and DeepSeek V4 Pro; the author declined to name a winner.

  Compared with kimi-k3 | 2026-10-04

  Practitioner report empiriolabs.ai