AI Index

GLM-5.3-Flash

Language Zhipu AI Verified Released Aug 26, 2026
  • In TextImageVideo Out Text
  • Reasoning· low, high, max
  • Tool use
  • Structured output
  • Open weights

Prices

5 offerings · prices as listed
Z.ai API Official Global · USD glm-5.3-flash
Conditions/ 1M tokensInputCache readOutput
Standard $0.15$0.030$0.50

Prices from Pricing (checked Nov 1, 2026)

Volcengine Ark Official China · CNY glm-5.3-flash
Conditions/ 1M tokensInputCache readOutput
Standard ¥0.80¥0.23¥2.80

Prices from 火山方舟：模型价格 (checked Nov 2, 2026)

Z.ai API Official China · CNY glm-5.3-flash
Conditions/ 1M tokensInputCache readOutput
Standard ¥0.80¥0.23¥2.80

Prices from 智谱 BigModel：API 定价 (checked Nov 2, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Reasoning can be off
—
Licence
mit
Family
glm-flash
Model ID

Not stated on Zhipu AI’s pages: max input, max thinking budget, built-in tools, knowledge cutoff, aliases.

Run locally

Weights: zai-org/GLM-5.3-Flash
  • Parameters 321.3B
  • Active 18B · 288 experts, 8 per token
  • Native precision BF16
  • Layers 45
  • KV cache per token 45 KB
  • Config positions 1024K (config.json max_position_embeddings)
  • API context 977K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Doesn’t fit IQ4_XS (automatic) 8K context
WeightsKV cacheOverheadUsable
Needs
≈ 167.3 GB (weights 159.0 + KV cache 0.4 + overhead 7.9)
Usable
21.6 GB (90% of its memory)
Try
it’s too large for this hardware; see the smallest that runs it below
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file305.8 GB≈ 321.4Doesn’t fit—
Q8_0318.0 GB≈ 334.2Doesn’t fit—
FP8299.3 GB≈ 314.6Doesn’t fit—
Q6_K245.8 GB≈ 258.4Doesn’t fit—
Q5_K_M212.8 GB≈ 223.8Doesn’t fit—
Q5_0205.7 GB≈ 216.4Doesn’t fit—
Q4_K_M181.8 GB≈ 191.2Doesn’t fit—
Q4_0168.3 GB≈ 177.1Doesn’t fit—
IQ4_XS159.0 GB≈ 167.3Doesn’t fit—
Q3_K_Mquality drops146.3 GB≈ 153.9Doesn’t fit—
IQ3_XXSquality drops114.5 GB≈ 120.5Doesn’t fit—
Q2_Kquality drops112.2 GB≈ 118.2Doesn’t fit—
IQ2_XXSquality drops77.1 GB≈ 81.3Doesn’t fit—

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): no single listed GPU; unified memory Apple M5 Ultra (64-core GPU) · 256 GB.

Run commands

The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.

Memory to fine-tune
Full≈ 4790.7 GBDoesn’t fit your hardware
LoRA≈ 601.5 GBDoesn’t fit your hardware
QLoRA≈ 171.3 GBDoesn’t fit your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • Epoch Capabilities Index 153.6

    #49 of 118 listed models · CI 151.9–155.5

  • LMArena Text 1,477

    #27 of 163 listed models · CI 1,472–1,481 · 41,642 votes

  • Epoch AI · Chess Puzzles 21%

    #60 of 85 listed models · max effort · ±4.09

  • Epoch AI · FrontierMath Tier 4 v2 17.1%

    #49 of 65 listed models · max effort · ±5.95

  • Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Oct 28 – Nov 5, 2026
  • LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 19 – 28, 2026
  • LiveBench (White et al.), release 2026-06-25, livebench.ai, licensed CC BY-SA 4.0. This board, adapted from it, is shared under CC BY-SA 4.0. · License · as of Jun 25, 2026
  • Terminal-Bench 4.0 leaderboard (Marten, Shaw et al., Terminal-Bench; harbor-framework/terminal-bench on GitHub), licensed Apache-2.0. · License · as of Sep 29, 2026
  • changes we made

Provider-reported scores

Published by Zhipu AI, under the conditions it states
BenchmarkSettingScore
Terminal-Bench 2.1Claude Code 2.1.207 harness, 6h timeout84.3%
Humanity's Last Examfull set, with tools, context management, 300K max context55.3%

Included in coding plans

All coding plans →

News

All news →

Changes

All changes →
  1. Now priced on Cloudflare Workers AI Input $0.15 · Cache read $0.030 · Output $0.50
  2. Now priced on SiliconFlow, China Input ¥0.80 · Output ¥2.80 · Cache read ¥0.23
  3. New language model

Sources

Zhipu AI’s official policies: data use, privacy, terms →

Zhipu AI · China · z.ai

Snapshots, the same family and other models from this provider