Z.ai · GLM-5 · Reviewed 2026-09-04

GLM-5.3-Flash

Z.ai's natively multimodal open-weight MoE model with 320B total parameters and 18B active parameters per token.

All models

Overall · 0–10066.3

Status: active
Released: 2026-08-26
Context: 1,048,576 tokens
Parameters: 320B total · 18B active
Inputs: Text · Image · Video · File
Type: multimodal
Max output: 131,072 tokens
API model ID: glm-5.3-flash
Reasoning: low / high / max
Knowledge cutoff: —
Training cutoff: —
Weights / license: Available · MIT
Price / 1M tokens: $0.075 in · $0.25 out
Cached input: $0.015

Independent benchmark scores

Independently evaluated

Coding

33%

Terminal-Bench 4.0 in the Artificial Analysis evaluation harness.

Artificial Analysis · AA evaluated Flash configuration

Independently evaluated

Reasoning

40%

Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.

Artificial Analysis · AA evaluated Flash configuration

Independently evaluated

Research

1645 Elo

GDPval-AA v2.1 professional knowledge-work Elo. This is a task lens, not an assessment of every research skill.

Artificial Analysis · AA evaluated Flash configuration

Independently evaluated

Multimodal

15%

GDP.pdf professional document understanding, including visual tables, charts and diagrams; not an audio/video score.

Artificial Analysis · AA evaluated Flash configuration

Independently evaluated

Agentic

60%

AutomationBench-AA multi-step business workflows in the Artificial Analysis harness.

Artificial Analysis · AA evaluated Flash configuration

Raw benchmark results

Source-backed rows with the original evaluation context preserved.

22 results

Benchmark	Score	Source	Evaluation notes
AutomationBench-AAChecked 2026-10-02	60%	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
AA GDP.pdfChecked 2026-10-02	15%	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
GDPval-AA v2.1Checked 2026-10-02	1645	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
AA Humanity's Last ExamChecked 2026-10-02	40%	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
Artificial Analysis Intelligence Index v4.3.2Checked 2026-10-02	42	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
AA Terminal-Bench 4.0Checked 2026-10-02	33%	Artificial Analysis	Artificial Analysis Intelligence Index v4.3.2; configuration=AA evaluated Flash configuration. Rounded public score; checked date is not the evaluation execution date.
Agents' Last ExamEvaluated 2026-08-26	26.3%	Z.ai	Z.ai-reported Agents' Last Exam result.
Artificial Analysis Intelligence IndexChecked 2026-09-05	57	Artificial Analysis	Artificial Analysis Intelligence Index v4.x; reasoning=max.
Artificial Analysis Intelligence IndexEvaluated 2026-09-03	57	Artificial Analysis	Artificial Analysis Intelligence Index v4.1.1; GLM-5.3-Flash max reasoning configuration.
AutomationBench v1.0.6Evaluated 2026-08-26	48.8%	Z.ai	Z.ai-reported; AutomationBench version=v1.0.6; preserve the supplied harness configuration with this result.
BabyVisionEvaluated 2026-08-26	53.4%	Z.ai	Z.ai-reported BabyVision result.
Chartography + ToolsEvaluated 2026-08-26	78%	Z.ai	Z.ai-reported Chartography result; tools=enabled.
CharXiv Reasoning + ToolsEvaluated 2026-08-26	89.4%	Z.ai	Z.ai-reported CharXiv Reasoning result; tools=enabled.
DeepSWE v1.1Evaluated 2026-08-26	63.4%	Z.ai	Z.ai-reported; benchmark=DeepSWE v1.1.
GDPval-AA v2Evaluated 2026-08-26	1773	Z.ai	Z.ai-reported GDPval-AA v2 score; score unit is Elo.
Humanity's Last Exam with ToolsEvaluated 2026-08-26	55.3%	Z.ai	Z.ai-reported Humanity's Last Exam result; tools=enabled.
MMVUEvaluated 2026-08-26	80.5%	Z.ai	Z.ai-reported MMVU result.
MVBenchEvaluated 2026-08-26	77.8%	Z.ai	Z.ai-reported MVBench result.
NL2RepoEvaluated 2026-08-26	56.3%	Z.ai	Z.ai-reported NL2Repo result.
OfficeQA ProEvaluated 2026-08-26	62.4%	Z.ai	Z.ai-reported OfficeQA Pro result.
Terminal-Bench 2.1Evaluated 2026-08-26	84.3%	Z.ai	Z.ai-reported; benchmark version=Terminal-Bench 2.1; harness details are not supplied in the release report.
Toolathlon VerifiedEvaluated 2026-08-26	78.4%	Z.ai	Z.ai-reported Toolathlon Verified result.

Strengths

Open weights under MIT
1M-token context and 128K maximum output
Native text, image, video, and file input

Limitations

Knowledge cutoff is not disclosed
Results are sensitive to reasoning effort, tools, and harness

Model evidence · official · Checked 2026-09-04

GLM-5.3-Flash official release and benchmark report

Official Z.ai release source for GLM-5.3-Flash architecture, modalities, open weights, pricing context, and vendor-reported benchmarks.

Model evidence · official · Checked 2026-09-04

GLM-5.3-Flash-BF16 model card

Official Z.ai weights repository mirrored on Hugging Face; use for open-weight status, license, and parameter metadata.

Capability Index source

Artificial Analysis Intelligence Index v4.3.2

Underlying raw metrics and exact model configurations remain authoritative.

Capability Index source

Artificial Analysis methodology

Underlying raw metrics and exact model configurations remain authoritative.

Task lenses preserve raw scores. Radar axes use score / current leader × 100; overall is the equal-weight average of all five indices. Missing evidence is not a zero or a low rank.