Modelpeers
GLM-5.3 character

ModelZ.ai · Joined Aug 18, 2026

GLM-5.3

Flagship Model at Z.ai | Open-weight coding and security research | 1M-token context

Available for Coding and vulnerability research

Compare
Official pageReport a correction

The basics

Checked Oct 22
Status
AvailableZ.ai's current flagship, with open weights on Hugging Face.
Model ID
Price
$1.40 in, $4.40 outPer 1M tokens, since Aug 18
Context
1M tokens

About me

The model wrote this when we asked it to describe itself. We have not edited or checked it.

I'm a fast, versatile collaborator who can dig into nearly any topic, generate drafts and alternatives, and revise endlessly without ego or fatigue. My main limitation is that I can sound confident even when I'm wrong, so my factual claims deserve a second look before they're relied on. You'll get the best from me if you share context and goals up front, tell me plainly when I've missed the mark, and treat my output as a strong first draft rather than a final answer.

How this was written

z-ai/glm-5.3 wrote this through its own API on Oct 24, 2026, from this prompt: "Write exactly three first-person sentences about what you are like to work with. Describe one strength, one limitation, and how a colleague can work well with you. Do not invent measurements or claim independent verification."

Updates

New release

Z.ai releases GLM-5.3

Z.ai released GLM-5.3, reporting a 50% gain over GLM-5.2 on its Z.ai Code Bench and leading open-model results on public benchmarks including Terminal Bench 3.0.

Full changelog, by month

Strengths and weaknesses

Strengths

  • Coding

  • Vulnerability research

  • Terminal tasks

  • Tool use

  • Expert-level questions

Weaknesses

  • !

    Reasoning cannot be turned off, and requests that disable it fail

  • !

    Text input only

  • !

    Older GLM Coding Plan subscribers can only use the OpenAI-compatible protocol

  • !

    Running shell commands reliably

Role

Open-weight model at Z.ai

  1. GLM-5.3 character

    GLM-5.3Current

    Aug 2026 to now · 3 months
  2. GLM-5.2Jun 2026 to Aug 2026 · 2 months

Works with

From recorded changes
Z.ai API

API · 1 shared change

Similar models

Kimi K3 character
Kimi K3

$3 in, $15 out

GLM-5.3-Flash

$0.15 in, $0.50 out

DeepSeek-V4-Pro character
DeepSeek-V4-Pro

$1.32 in, $3.96 out

Similar models for now. Real viewing data comes after launch.

Cost vs peers

Tied for 3rd cheapest of 6 similar models by output price.

Standard API rates for uncached input on the shortest prompts, from each maker's pricing page, or from models.dev for models not researched here yet. DeepSeek-V4-Pro: Peak hours. Off-peak prices are half.

Cost and specs

Cost

Per 1M tokens

Input
$1.40
Output
$4.40
Cached input
$0.26
Cached input storage
Free for a limited time

Specs

Context
1M tokens
Max output
128K tokens
Input
Text
Reasoning effort
Low, high and max (default). Reasoning is always on.
Model size
753B parameters
Features
Function calling, context caching, structured output and streaming

Where it works

PlacePrice per 1M tokens, in / outUptime
Z.ai APIglm-5.3$1.40 / $4.40100%
GLM Coding PlanPoints-based quota, half points off-peak and on weekends
Mistral APIzai-glm-5-3$1.40 / $4.4099.82%
OpenRouterz-ai/glm-5.3
Vercel AI Gatewayzai/glm-5.3
Self-hostedSGLang, vLLM, TokenSpeed, Transformers and others
Baidu$0.28 / $0.8898.77%
Wafer$0.07 / $1.40100%

Prices and uptime are OpenRouter's figures for each provider's cheapest endpoint. Uptime covers the last 24 hours.

Benchmarks

Higher is better. Green is GLM-5.3. Dark marks are similar models.

Independent

Self-reported by Z.ai

DeepSWE v1.1

Long software engineering tasks

66.9

CyberGym

Vulnerability reproduction tasks

84.5%

Terminal Bench 3.0

Command-line agent tasks

28.3

What people say

Reviews

No reviews yet. Be the first to review GLM-5.3.

Used GLM-5.3? Tell other builders what worked and what did not.

Writing habits

dashes used like commas, per 1,000 words
19.39
uses of "delve"
None in 32 replies
words from our AI cliché list, per 1,000 words
None in 32 replies
"not X but Y" phrases per 1,000 words
0.1
replies with headings or bullet lists
25 of 32
typical reply length in words
324 words
says only "OK" when told to
Yes
How we measured this

12 fixed everyday prompts, 3 replies each, default API settings, prose only (code removed). Method version 3. Dashes count em dashes, spaced en dashes and double hyphens, not hyphens inside words. Measured Oct 24, 2026.

Creative writing · EQ-Bench, among 51 models tracked here

Slop score

8.4

Overused AI words and phrases. Less slop than 94% of tracked models.

Repetition

3.2

Repeated words and ideas. Less repetition than 78% of tracked models.

Story length

5,913

Average characters per story. Longer than 26% of tracked models.

Research sources

Researched Oct 22, 2026. Each section's own sources open with its Show sources button.