GLM-5.3-Flash

Frontier coding and multimodal intelligence engineered for flash-level inference cost

Try now

read docs

About model

GLM-5.3-Flash is the first natively multimodal model in Z.ai's GLM-5 series, built to deliver frontier coding and agentic capability at a fraction of typical inference cost. A 320B parameter Mixture-of-Experts model with just 18B active, it introduces a hybrid architecture combining linear and sparse attention with a lightweight indexer, sharply reducing long-context serving cost across its 1M-token window while preserving long-context precision. Trained on a 30T-token multimodal corpus, the model brings vision into the coding loop, inspecting its own rendered outputs and refining them iteratively, and reads the documents, spreadsheets, dashboards, and interfaces that professional work runs on. Previewed anonymously as Ox Alpha before launch, with weights publicly available. Available on Together AI.

‍

Total Parameters (18B active)

320B

Nearly half the active parameters and layers of GLM-4.5 at a similar total size

Context Window

Hybrid linear and sparse attention built to keep long context cheap

Multimodal Input

Native

Vision integrated into the coding loop and professional document work

Model key capabilities

Frontier Coding at Flash Cost: Coding and agentic gains far beyond GLM-5.2, from an architecture designed to produce more intelligence with less compute
Visual Intelligence in the Coding Loop: Inspects its own rendered outputs and refines them iteratively, extending validation beyond functional correctness to the interfaces users actually see
Efficient Long Context: Hybrid linear and sparse attention with IndexPool, cutting attention compute 3.0x and KV cache 4.4x versus GLM-5.3 across a 1M-token window
Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

API usage

cURL
Python
Typescript

Endpoint:

zai-org/GLM-5.3-Flash

curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.3-Flash",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="zai-org/GLM-5.3-Flash",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'zai-org/GLM-5.3-Flash',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• Mixture-of-Experts model with 320B total parameters, 18B active, and 45 layers, nearly halving both the active parameter count and layer depth of GLM-4.5 at a similar total size
• Hybrid attention: linear attention captures local dependencies through state modeling while sparse attention retrieves global context through a lightweight indexer
• IndexPool compresses four indexer key vectors into one through weighted pooling, reducing indexer latency and memory at 1M-token context
• Manifold-Constrained Hyper-Connections (mHC) improve scaling efficiency; Multi-Token Prediction layer included
• Native vision encoder processing images alongside text; 1M-token context window
• Compared with GLM-5.3, attention compute is reduced 3.0x and KV cache size 4.4x per Z.ai's measurements

Training Methodology:
• Pre-trained on Z.ai's latest 30T-token multimodal corpus
• Data synthesis pipelines for visual coding focused on self-visual judgment and test-time improvement, requiring the model to interact with environments, inspect outputs, and refine iteratively
• Reinforcement learning with environment feedback for frontend coding, with GUI judgment strengthened through agent-based verification grounded in real user flows
• Previewed anonymously as Ox Alpha to gather real-world user feedback before launch

Performance Characteristics:
• Terminal coding (Terminal Bench 2.1, completing real tasks in a command-line environment): 84.3, up from 81.0 for GLM-5.2
• Repository-scale engineering (DeepSWE v1.1): 63.4, up from 46.2; repository generation (NL2Repo): 56.3
• General agent tasks (Toolathlon Verified multi-tool workflows: 78.4, up from 59.9; AutomationBench: 48.8, up from 26.2; Agents' Last Exam: 26.3; Humanity's Last Exam with tools: 55.3)
• Knowledge work (GDPval-AA v2 Elo, evaluated by Artificial Analysis): 1773; Artificial Analysis Intelligence Index v4.1.1: 57
• Vision (OfficeQA Pro document analysis: 62.4; CharXiv Reasoning with tools: 89.4; Chartography with tools: 78.0; BabyVision general visual reasoning: 53.4; MVBench: 77.8; MMVU: 80.5)
• Z.ai reports the model outperforms GLM-5.2 at every effort level on its in-house Z.ai Code Bench while using fewer output tokens
‍
Prompting
Together AI API Access:
• Access GLM-5.3-Flash via Together AI APIs using the endpoint zai-org/GLM-5.3-Flash
• Authenticate using your Together AI API key in request headers
• Set reasoning effort to low, high, or max per request to trade response speed against deliberation depth
• Supports text and image input across a 1M-token context window
• Available on Together AI serverless and dedicated infrastructure
‍
Applications & use cases
Cost-Efficient Agentic Coding:
• Run high-volume coding agents with an 18B-active-parameter cost profile
• Hold entire repositories in the 1M-token window across extended sessions
• Scale concurrent agent workloads where per-request cost matters as much as capability

Visual Coding & Frontend Work:
• Build frontend, game, and simulation projects where the model verifies its own rendered output
• Ground GUI work in visual feedback rather than code-only validation
• Iterate on interfaces with self-inspection driving each refinement pass

Professional Document & Chart Work:
• Reason over documents, spreadsheets, dashboards, and presentations directly
• Run chart analysis and data-extraction workflows through native vision input
• Assess outputs against visual context for presentation quality, not just correctness
‍

Related models

Model specifications

Model data

Model provider
ZAI
Type
Chat
Reasoning
Vision
Deployment
Serverless
Dedicated
Endpoint
zai-org/GLM-5.3-Flash
Parameters
320B
Activated parameters
18B
Context length
1M
Input price
$0.15 / 1M tokens
$0.03 (cached)/1M
Output price
$0.50 / 1M tokens
Input modalities
Text
Image
Output modalities
Text

Released
August 25, 2026
External link
Provider docs
Category
Chat

Run in Playground

Quickstart docs

Deploy model

GLM-5.3-Flash

About model

API usage

Model card

Prompting

Applications & use cases