GLM-5.3-Flash
A fast, multimodal reasoning model built for coding, tool use, and long-running agentic workflows.
Quantization
FP8
320B MoE · 18B active
Context Window
1.0M
Tokens
Intelligence
57
AA Index
Released
Aug 2026
26 Aug 2026
Frontier agentic quality, Flash-tier economics
GLM-5.3-Flash is Z.ai's natively multimodal model for coding, reasoning and agentic workflows. It has 320 billion total parameters with 18 billion active per token, combining sparse and linear attention for efficient long-context processing.
Trained on a 30-trillion-token multimodal corpus, it improves on GLM-5.2 across coding, tool-use and general capability benchmarks while requiring substantially less serving compute — which is why it lands at a fraction of the price of comparable closed frontier models.
320B MoE · 18B active per token
Best for long-running, tool-heavy agents
Agentic Workflows
Built for long-running agents: sustained planning, tool orchestration and recovery across hundreds of steps.
Software Engineering
Improves on GLM-5.2 across coding benchmarks — multi-file edits, repo-scale refactors and iterative debugging.
Multimodal Understanding
Natively multimodal: screenshots, diagrams, dashboards and documents alongside text in the same request.
Long-Context Analysis
Sparse plus linear attention keeps 1M-token context efficient for large codebases and document sets.
Also strong on complex reasoning and multimodal document analysis.
Flexible Pricing Tiers
Choose the optimal balance of speed and cost for your workflow. Prices are per 1M tokens.
| Tier | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Batch | $0.08 | $0.25 |
| Async | $0.11 | $0.38 |
| Realtime | $0.15 | $0.50 |
Context window natively supported up to 1,048,576 tokens. Custom pricing available for bulk discounts and large workloads — reach out to hello@doubleword.ai.
Start Building in Minutes
GLM-5.3-Flash is accessible via OpenAI-compatible endpoints. Default sampling: temperature=1.0, top_p=0.95.
from openai import OpenAI
client = OpenAI(
api_key="your-api-key-here",
base_url="https://api.doubleword.ai/v1"
)
# Multimodal, long-running agentic task
response = client.chat.completions.create(
model="zai-org/GLM-5.3-Flash",
messages=[
{"role": "user", "content": "Trace this stack trace screenshot back to the failing module and patch it."}
],
temperature=1.0,
top_p=0.95,
)
print(response.choices[0].message.content)💡 Pro Tip
Route long-running agent steps through the async tier and bulk evals through batch — same model, same API, up to 47% off the realtime rate. See the docs for the batch job format.