GLM-5.3-Flash

Across the task benchmarks we track, GLM-5.3-Flash’s strongest relative results are in terminal use, work automations, and knowledge work. On Terminal-Bench 2.1, it scores 84.3%, ranking #2 among 13 models with reported results in our dataset.

Updated: Nov 5, 2026Data as of: Nov 2, 2026

Compare this model

Strongest results

Strongest results for GLM-5.3-Flash in tracked evaluation cohorts
Benchmark	Measures	Score
Agent Index ↗v1.2 Preview	Overall agent capability	23.3 / 100#20 of 26
AutomationBench-AA ↗1.0.6 / AA private 657	Work automations	60.37%#15 of 55
GDPval-AA ↗2.1	Knowledge work	1,649.37 Elo#11 of 38
AA-Briefcase ↗1.1 / combined Elo	Business deliverables	1,450.23 Elo#20 of 57

Model details

Open source / open weights19 benchmarks

Specifications for GLM-5.3-Flash
Model	GLM-5.3-Flash
Agent Index v1.2	23.3 / 100 · Rank 20 · Preview
Creator	Z.ai (opens in a new tab)
Availability	Open source / open weights
Context size	1,000,000 tokens (opens in a new tab)
Max output	Not reported
Knowledge cutoff	Not reported
License	MIT · Model weights ↗ (opens in a new tab)

Pricing and performance

Per provider and reported configuration. Prices are USD per 1M tokens.

Provider offerings for GLM-5.3-Flash
Provider / configuration	Contexttokens	InputUSD / 1M tokens	OutputUSD / 1M tokens	LatencyTTFT · seconds	Speedtokens / sec
Z.ai (opens in a new tab) GLM-5.3-Flash	1,000,000	$0.15 (opens in a new tab)	$0.50 (opens in a new tab)	1.546 (opens in a new tab)	75.174 (opens in a new tab)

Benchmark performance

GLM-5.3-Flash in color, alongside leading models. Latest stored reports, best effort per report date.

Reasoning

198-question Diamond subset; independent AA evaluation

GPQA Diamond ↗

01
GPT-6 Astra96.3%
02
Gemini 3.8 Flash95.3%
03
Gemini 3.1 Pro94.1%
03
GPT-5.6 Sol94.1%
03
Muse Spark 1.394.1%
16
GLM-5.3-FlashThis model91.2%

Terminal Use

89 tasks; Terminus 2; e2b; three repeats

Terminal-Bench 2.1 ↗

01
Kimi K385%
02
GLM-5.3-FlashThis model84.3%
03
GLM-5.383.9%
04
DeepSeek V4 Pro 081378.7%
05
GLM-5.277.9%

Work Automations

Private 657 tasks; 50 turns; objective completion

AutomationBench-AA ↗

01
Gemini 4 Carbon77.8%
02
Gemini 4 Argon77.5%
03
Kimi K3.1 Pro73.3%
04
MiniMax M3.1 Flash72.52%
05
Claude Sonnet 5.571.8%
15
GLM-5.3-FlashThis model60.37%

Knowledge Work

220 tasks; Stirrup agent; pairwise Elo

GDPval-AA v2.1 ↗

01
Claude Opus 5.51,866 Elo
02
Kimi K3.1 Pro1,858 Elo
03
Claude Sonnet 5.51,838 Elo
04
Claude Fable 5.11,763 Elo
05
DeepSeek V4.1 Flash 10171,746 Elo
11
GLM-5.3-FlashThis model1,649 Elo

Reported benchmarks

All published results in our dataset, with versions, evaluation settings, and sources.

All published benchmark reports for GLM-5.3-Flash
Benchmark	Score	Evaluation setup	Source
AA-Briefcase1.1 / combined EloCombined Elo · Higher is better	1,450.23 Elo	selection: best reported configuration per release · GLM-5.3-Flash · tools: benchmark-defined	Artificial Analysis (opens in a new tab)independent
AutomationBench1.0.6 / Z.ai provider reportStrict task completion · Higher is better	48.8%	harness comparison: provider-reported; settings differ across models · bench_53.png; v1.0.6 with null-type fix PR13; exact task set not established	Z.ai ↗ (opens in a new tab)provider reported
AutomationBench-AA1.0.6 / AA private 657Objectives completed without guardrail violations · Higher is better	60.369%	GLM-5.3-Flash; Private 657 tasks; 50 turns; objective completion	Artificial Analysis (opens in a new tab)independent
GDP.pdf100 tasks / All-pass / five attemptsAccuracy · Higher is better	15.4%	selection: best reported configuration per release · GLM-5.3-Flash · tools: benchmark-defined	Artificial Analysis (opens in a new tab)independent
GDPval-AA2.1Elo · Higher is better	1,649.37 Elo	GLM-5.3-Flash; 220 tasks; Stirrup agent; pairwise Elo	Artificial Analysis (opens in a new tab)independent
GPQADiamond / Artificial AnalysisAccuracy · Higher is better	91.212%	GLM-5.3-Flash; 198-question Diamond subset; independent AA evaluation	Artificial Analysis (opens in a new tab)independent
HLE-Diamond1,000 questions / no toolsAccuracy · Higher is better	18.2%	selection: best reported configuration per release · GLM-5.3-Flash · No tools	Artificial Analysis (opens in a new tab)independent
Humanity's Last ExamStable HLE — full setAccuracy · Higher is better	55.3%	Primary bench_53.png; full set; temperature 1; top_p .95; 163840 generation; 300K context; GPT-5.6 Luna judge · Tools enabled	Z.ai ↗ (opens in a new tab)provider reported
Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better	39.852%	GLM-5.3-Flash; original HLE text-only 2,158 questions; no tools; not HLE Diamond · No tools	Artificial Analysis (opens in a new tab)independent
LiveBench: Agentic2.0 / 2026-10-25Accuracy · Higher is better	19%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: agentic	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Browser Use2.0 / 2026-10-25Accuracy · Higher is better	59.8%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: browser-use	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Coding2.0 / 2026-10-25Accuracy · Higher is better	80.6%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: coding	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Data Analysis2.0 / 2026-10-25Accuracy · Higher is better	76.4%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: data-analysis	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Design2.0 / 2026-10-25Accuracy · Higher is better	77.2%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: design	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Reasoning2.0 / 2026-10-25Accuracy · Higher is better	77.6%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: reasoning	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
LiveBench: Research2.0 / 2026-10-25Accuracy · Higher is better	70.1%	release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: research	LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25
Terminal-Bench2.1 / Artificial AnalysisPass@1 · Higher is better	84.27%	GLM-5.3-Flash; 89 tasks; Terminus 2; e2b; three repeats	Artificial Analysis (opens in a new tab)independent
Terminal-Bench2.1 / published open-model reportsTask success · Higher is better	84.3%	harness comparison: provider-reported; settings differ across models · Claude Code 2.1.207; 6h; 65536 output tokens	Z.ai ↗ (opens in a new tab)provider reported
Terminal-Bench4.0 / mini-swe-agentTask success · Higher is better	32.828%	harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GLM-5.3-Flash	Artificial Analysis (opens in a new tab)independent