GLM-5.3

By Z.ai. 5 results on 5 boards, copied as published. Different boards, different metrics: don't add them up.

License (as listed)MIT

Hosting regionNot confirmed

Boards5

Code

95.4%Vals AI — SWE-bench Verified
GLM 5.3 · % resolved · board #6
Mini-SWE-agent · as of 2026-09-01 Stale ↗
80.53%Vals AI — LiveCodeBench
GLM 5.3 · Accuracy % · board #68
Vals AI implementation (no agent) · as of 2026-09-01 Stale ↗

Agentic / terminal

41.82%Terminal-Bench 4.0
GLM-5.3 (max) · Resolution rate
Claude Code · as of 2026-08-14 ↗

Websites

1623LMArena WebDev Arena
glm-5.3-max · Arena score · board #25
LMArena WebDev (agentic front-end tasks) · as of 2026-10-22 ↗
78.13%Vals AI Vibe Code Bench v1.1
GLM 5.3 · Score % · board #30
OpenHands · as of 2026-10-20 ↗

Design

No result on the boards we captured

Android

No result on the boards we captured

iOS

No mature public leaderboard

Available in

No plan row in our pricing data names this model version.

Use cases with this model

GLM-5.3

GLM-5.3 one shot, make a minecraft clone

Builder: @0xSero
Posted: Oct 13, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

GLM-5.3

Living Weights ported to GLM-5.3 (753B): "me and my coding agent"

Builder: @volatilemarkts
Posted: Oct 10, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

Kimi K3GLM-5.3Qwen 3.8 Max+2Model test

Test: "Qwen 3.8 Max (max): 25.7 - DeepSeek V4.1 Flash (max): 21.7 - Kimi K3 (default): 23 - GLM-5.3 (max): 19 - Mistral Large 4 (high, its max): 15"

Builder: @PawelHuryn
Posted: Oct 7, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

Kimi K3GLM 5.3Model test

"benchmarked worse + similar price as Kimi K3 and GLM 5.3 in the Pi harness", run on real PRs

Builder: @avyvar
Posted: Oct 7, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

DeepSeek V4 ProGLM 5.3GPT-6 Astra+5

Same 3D prompt one-shot across 8 models incl. Mistral Large 4

Builder: @domsilvestre
Posted: Oct 6, 2026
Plan: —
Cost: Not disclosed

DesignRead the post

x.com · link re-checked 2026-10-16

Clineglm-5.3

Plan: "flat $18/mo for glm-5.3 access which hooks directly into cline, roo code, and cursor"

Builder: @mzulfaas
Posted: Oct 4, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

GLM-5.3Model test

Test: a Blender battle, DeepSeek-v4.1 vs GLM-5.3-Flash, building a "Dodecahedron trapped inside a dodecahedron"

Builder: @WescheNex1q
Posted: Sep 18, 2026
Plan: —
Cost: Not disclosed

DesignRead the post

x.com · link re-checked 2026-10-16

Same family: GLM 8

Factoryglm 5.3 flash

Complaint: "Very Slow Workers … i was using very fast models like glm 5.3 flash"

Builder: @willyoudeal
Posted: Oct 3, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

GLM 5.3 Flash

Spent September on one open model: 2B tokens; the vibe-coded Wagtail MCP server prototype

Builder: Thibaud Colas
Posted: Oct 2, 2026
Plan: GLM 5.3 Flash
Cost: Stated: $68" for GLM 5.3 Flash usage; MCP prototype "450M tokens / $150

CodeRead the post

wagtail.org · web source · link re-checked 2026-10-16

GLM-5.3 Flash

Voice AI agent booking appointments over live phone calls — "Runs on GLM-5.3 Flash through Telnyx, about $0.30 per booking call"

Builder: @Sumanth_077
Posted: Sep 30, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

GLM 5.3 Flash

I'm back on GLM 5.3 Flash. On 2x Sparks w/ZCode it works so well … ZCode isn't open source and has some telemetry

Builder: @imost
Posted: Sep 16, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

ClineDeepSeek V4 FlashDeepSeek V4 ProGLM 5.3 Flash

Cline Desktop + DeepSeek V4 Pro: "pointed it at an empty folder, and asked it to build an evaluation harness for an ML…"

Builder: @akshay_pachaar
Posted: Sep 14, 2026
Plan: —
Cost: Not disclosed

CodeRead the post

x.com · link re-checked 2026-10-16

All 8 GLM use cases →

Sources

lmarena.ai/leaderboard/webdev vals.ai/benchmarks/vibe-code vals.ai/benchmarks/swebench vals.ai/benchmarks/lcb tbench.ai/leaderboard

GLM-5.3

Available in

Use cases with this model

GLM-5.3 one shot, make a minecraft clone

Mainbrella Build: a Lovable-style app builder running open models on Cloudflare Workers AI

Living Weights ported to GLM-5.3 (753B): "me and my coding agent"

Test: "Qwen 3.8 Max (max): 25.7 - DeepSeek V4.1 Flash (max): 21.7 - Kimi K3 (default): 23 - GLM-5.3 (max): 19 - Mistral Large 4 (high, its max): 15"

"benchmarked worse + similar price as Kimi K3 and GLM 5.3 in the Pi harness", run on real PRs

Same 3D prompt one-shot across 8 models incl. Mistral Large 4

Self-playing Flappy Bird and Missile Command games, each one-shot in a single HTML file, comparing 4 open models

Plan: "flat $18/mo for glm-5.3 access which hooks directly into cline, roo code, and cursor"

Test: a Blender battle, DeepSeek-v4.1 vs GLM-5.3-Flash, building a "Dodecahedron trapped inside a dodecahedron"

Same family: GLM 8

Complaint: "Very Slow Workers … i was using very fast models like glm 5.3 flash"

Spent September on one open model: 2B tokens; the vibe-coded Wagtail MCP server prototype

Voice AI agent booking appointments over live phone calls — "Runs on GLM-5.3 Flash through Telnyx, about $0.30 per booking call"

Phone receptionist for a fictional café on GLM-5.3 Flash/Telnyx

I'm back on GLM 5.3 Flash. On 2x Sparks w/ZCode it works so well … ZCode isn't open source and has some telemetry

Cline Desktop + DeepSeek V4 Pro: "pointed it at an empty folder, and asked it to build an evaluation harness for an ML…"

Sources

The weekly diff