GLM-5.3

Name: GLM-5.3
Author: Zhipu AI

Zhipu AI

GLM-5.3 is Zhipu AI's 753B-parameter open-weight MoE model released in August 2026, with a 1M token context window. It is tuned for agentic workflows, showing strong results on Terminal-Bench 2.1, CyberGym and FrontierSWE. It is served by Novita and Z.ai.

Key Specifications

Parameters

753.0B

Context

1.0M

Release Date

August 14, 2026

Average Score

52.2%

API Documentation Research Paper Results Blog

Timeline

Key dates in the model's history

Announcement

August 14, 2026

Last Update

August 27, 2026

Today

September 8, 2026

Technical Specifications

Parameters

753.0B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$1.40

Output (per 1M tokens)

$4.40

Max Input Tokens

1.0M

Max Output Tokens

131.1K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Other Tests

Specialized benchmarks

Terminal-Bench 2.1

Claude Code 2.1.207; max reasoning effort • Self-reported

88.0%

CyberGym

Claude Code 2.1.207; max reasoning effort; single-run Pass@1 over 1,507 tasks • Self-reported

84.5%

FrontierSWE

Claude Code 2.1.207; max reasoning effort • Self-reported

78.0%

Agents' Last Exam

ALE-CLI; Claude Code harness; max reasoning effort; 1M-token context • Self-reported

28.5%

AutomationBench

AutomationBench v1.0.6 • Self-reported

48.2%

DeepSWE 1.1

DeepSWE v1.1; mini-swe-agent harness; max reasoning effort • Self-reported

66.9%

ExploitBench

Claude Code 2.1.207; max reasoning effort; average coverage over 41 tasks and 3 revisions • Self-reported

54.4%

ExploitGym

Six-hour normalized inference-time budget; Claude Code 2.1.207; max reasoning effort • Self-reported

15.0%

GDPval-AA

GDPval-AA v2 Elo (max_score 3000) • Self-reported

59.0%

Humanity's Last Exam

With tools; 300K-token context; GPT-5.6 Luna judge • Self-reported

62.5%

NL2Repo

1M-token context; rule-based and LLM-based anti-hacking checks • Self-reported

58.0%

PostTrainBench

Claude Code 2.1.207; max reasoning effort; weighted average of 3 runs • Self-reported

39.8%

Program Bench

Almost Solved metric • Self-reported

19.0%

SWE-Marathon

SWE-Marathon v1.1; Claude Code 2.1.207; max reasoning effort • Self-reported

42.5%

Terminal-Bench 3.0

Claude Code 2.1.207; max reasoning effort • Self-reported

28.3%

Terminal-Bench 4.0

Terminal-Bench 4.0 public leaderboard; resolution rate 41.8% ± 3.2%; Claude Code agent; max effort. • Verified

41.8%

Toolathlon

Toolathlon Verified • Self-reported

73.0%

License & Metadata

License

mit

Announcement Date

August 14, 2026

Last Updated

August 27, 2026

Compare GLM-5.3

All comparisons

vs GLM-5.2 vs GLM-5.3 Flash vs Grok 4.5 vs GPT-5.6 Luna vs Gemini 3.5 Flash-Lite vs DeepSeek-V4-Flash-Vision-Exp vs Grok 4.6 vs Claude Opus 5

Articles about GLM-5.3

Zhipu AI releases GLM-5.3 open weights with 'emergent' cyber capabilities

Zhipu AI released GLM-5.3's full open weights: a 744B MoE whose post-training-only gains include 'emergent' cyber skills and doubled ExploitBench scores.

August 30, 2026

6 min

Similar Models

All Models

GLM-4.5

Zhipu AI

355.0B

Best score:0.8 (GPQA)

Released:Jul 2025

Price:$0.60/1M tokens

GLM-5.2

Zhipu AI

753.0B

Best score:0.9 (GPQA)

Released:Jun 2026

Price:$1.40/1M tokens

GLM-4.5-Air

Zhipu AI

106.0B

Best score:0.8 (TAU)

Released:Jul 2025

GLM-5

Zhipu AI

744.0B

Best score:0.9 (TAU)

Released:Feb 2026

GLM-5.1

Zhipu AI

754.0B

Best score:0.9 (GPQA)

Released:Apr 2026

Price:$1.40/1M tokens

GLM-4.7

Zhipu AI

358.0B

Best score:0.9 (TAU)

Released:Dec 2025

Price:$0.60/1M tokens

GLM-4.7-Flash

Zhipu AI

30.0B

Best score:0.8 (TAU)

Released:Jan 2026

Price:$0.07/1M tokens

GLM-5.3 Flash

Zhipu AI

MM320.0B

Released:Aug 2026

Price:$0.15/1M tokens

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.