Z.ai
Z.ai

GLM 5.3 Flash

Release Date 2026-08-26(Knowledge cutoff 2026-02-17)1.3M IN · 131K OUTOpen Model(MIT)API availableEnds 2098-12-31

GLM 5.3 Flash is Z.ai's multimodal model for development and office workflows grounded in screens and documents. Unlike the text-focused standard model of its generation, it accepts images, video, and files; official guidance highlights inspecting interfaces, revising results with tools, and producing documents. AIB currently places automation in the top tier and vision and coding in the upper tier, while file creation and execution require a supporting tool environment. Its downloadable weights use MIT terms, and the official API keeps reasoning enabled, so latency and response length should be evaluated with the intended settings. Consider it for implementing interfaces from visual material or assembling documents, without importing the separate FlashX variant's speed or availability claims.

API cost

Market price (median)
$0.12IN$0.5OUT
Price range35 providers
$0.0399~$0.175IN$0.0784~$0.65OUT
IN
OUT
10/2311/22
SourceOpenRouterLast updated

Real-world usage

No.4Same as last week3.6% share · 10T tokens/wk
12T8/31
12T9/7
14T9/14
16T9/21
9.7T9/28
12T10/5
12T10/12
13T10/19
15T10/26
13T11/2
12T11/9
10T11/16
Top use cases
  • Running terminal commands17.4%
  • Code debugging15.2%
  • Editing code files13.5%
SourceOpenRouterLast updated

Output Speed

Fastper 2,000 tokens
105tok/s
First Output 0.6stotal: 19.6s
SourceOpenRouterLast updated

Expert Performance Evaluation

27.8147560.771.6

More about AA Intelligence Index v5

47.2%20.759.8

More about τ³-Banking

80.0%76.4147152.8%57.9%31.6%45.2%10.4%73.776.460.6%14.0%6.7%

More about Long Context Reasoning

146155.3%73.92.9%0.149

More about Arena Text Factuality

32.8%152656.830484.3%71.5161151.6%70.931.8%63.4%151579.00.0%30.8%18.1%36.8%

More about Terminal-Bench 4.0

145214791435147477.388.9%143952.8

More about Arena Writing, Literature & Language

Can it make accurate judgment calls across many work domains?Hanno Labs DecisionBenchTop 20%#1
74.1

More about Hanno Labs DecisionBench

39.9%15.4%150991.2%93.9%81.255.8%17.1%21.0%52.5%1493149577.677.691.0%65.8%21.0%8.0%

More about Humanity's Last Exam

129789.4%

More about Arena Vision

Specialized fields

14301457150115131490147656.1%98.8%84.0%57.763.4%82.6%87.5%80.4%78.4%

More about Arena Korean

OddBench

376Reasoning effort: High396

More about CSAT 2026

See each evaluation’s detail page for its source and last updateData corrections & removal requests