System Card← leaderboard
Zhipu AI

GLM Flash

#21of 28 · general

GLM-5.3 Flash

320Bparams · MoE
1Mcontext
30Ttraining tokens
$0.15 / $0.50per 1M in / out
2026-08-20released
weights ↗repo ↗
50100Coding 30 · #19 of 28Agents 32 · #21 of 29Cyber 22 · #14 of 24Engineering 22 · #19 of 26Math 23 · #21 of 26Health 50 · #10 of 24Legal 26 · #22 of 27Professional 34 · #18 of 28Writing 22 · #15 of 21 · one sourceInstruct 20 · #18 of 25Hallucination 37 · #16 of 26Multimodal 23 · #17 of 24
Coding30#19
Agents32#21
Cyber22#14
Engineering22#19
Math23#21
Health50#10
Legal26#22
Professional34#18
Writing22#15
Instruct20#18
Hallucination37#16
Multimodal23#17

History

20%40%60%80%100%Apr 2026Jul 2026GLM-4.7 Flash · $0.06/M in · 2026-01-19 · still served · OTIS Mock AIME 2024-2025 41.7% fractional score (0–1 scale)GLM-5.3 Flash · $0.0261/M in · 2026-08-20 · still served · OTIS Mock AIME 2024-2025 93.9% fractional score (0–1 scale)4.7 Flash
  • Zhipu AI
  • older → newer
$ / M input tokens2 of 176 priced models · price is a third-party claim, not a vendor quote

Usage

62.5TtokensAug 20 – Oct 6▼ 26% 7d
20T40T60T80TAug 20Sep 1Oct 1

Geography

China14.9%
United States13.8%
India5.9%
Germany5.0%
Japan3.9%
Brazil3.2%
Indonesia2.6%
France2.3%
United Kingdom2.2%
Singapore2.2%

Effort

GLM-5.3 Flash
score %253035$0.15$0.2$0.3$0.5$ / task · log scaleLowHighMax

Measured

≈ imputed · amber dot = lab-self-reported among the sources
GPQA Diamond88.3%#20Epoch AIVals.ai
HLE47.6%#13Artificial AnalysisLLM Stats
MMLU-Pro86.1%#7Vals.ai
LiveBench71.6%#19LiveBench
SWE-bench V92.0%#3Vals.ai
LiveCodeBench80.5%#9Vals.ai
SciCode51.6%#20Artificial AnalysisEpoch AI
CritPt15.4%#21Artificial AnalysisEpoch AI
Terminal-Bench62.9%#15Vals.ai
APEX Agents52.8%#14Epoch AI
GDPval57.2%#10Artificial Analysis
Arena1475#13Arena
AA Index41.8#16Artificial Analysis
Epoch Capabilities Index153.1#17Epoch AI

Neighbors

MiniMax Flash→ stronger agentDeepSeek Pro→ follows instructions better · pricierStep→ broader · pricierClaude Haiku→ stronger mathMiMo Flash→ less factual