GLM 5.3 API
GLM-5.3 is Z.ai's open-weights text model for coding and agentic work, offering a 1-million-token context, three thinking effort levels, and gains focused on multi-step tasks.
GLM-5.3 is Z.ai's text-in, text-out model built for demanding coding and agentic workloads. It supports context of up to 1 million tokens and provides a thinking mode with three reasoning effort levels: low, high, and max. Thinking can no longer be disabled, which is a breaking change for applications that previously ran with thinking switched off. The model also offers streaming messages for real-time responses, function calling for tool integration, context caching to optimize long conversations, and structured output in formats such as JSON.
It is designed for long-horizon coding agents, complex software engineering, multi-step CLI work such as refactors and CI triage, long-running agent loops, repo-scale work, and ML infrastructure diagnosis and optimization, as well as broader agentic tasks beyond coding. According to Z.ai, it is the most capable open-weights model for coding, with state-of-the-art results among open-source models on Terminal Bench 3.0 and Agents' Last Exam, and state of the art on CyberGym for vulnerability discovery. Z.ai reports that comparable work can be done at approximately 50,000 tokens instead of 120,000, and that gains concentrate on multi-step work rather than one-shot completions.
GLM-5.3 uses the same base model as GLM-5.2, a 743-billion parameter Mixture-of-Experts architecture, and all improvements come from post-training. Vendor-reported figures have not yet been independently replicated, and capabilities remain significantly lower than current closed frontier models. Cyber capability is strongest in the front end of the exploitation chain, while deeper exploitation and complete offensive and defensive workflows still need improvement. Open weights were delayed for approximately two weeks after launch for a safety review.
Updated:
MODEL ID
glm-5.3- Context
- 1.049M
- Max output
- 943.718K
- Recommended API
- messages
- Input
- text
- Output
- text

Highlights
- 1-million-token context window
- Three thinking effort levels: low, high, max
- Function calling and structured JSON output
- Context caching for long conversations
- Text input and text output only
Providers
Best priceCompare API prices across providers. Select a provider to view all charges and conditions.
| Uptime | ||||
|---|---|---|---|---|
| $1.4 | $4.4 | $0.26 | ||
| $0.7 | $2.2 | $0.13 | ||
| $0.91 | $2.86 | $0.169 | ||
| $1.053 | $2.15 | $0.356 | ||
| $1.19 | $3.74 | $0.238 | ||
| $1.4 | $4.4 | $0.14 | ||
| $1.4 | $4.4 | $0.26 | ||
| $1.4 | $4.4 | $0.26 | ||
| $1.68 | $5.28 | $0.312 |
Historical observations are not a live health check. Missing data does not mean an outage.
What $10 buys
- Input
- 7.1M tokens
- Output
- 2.3M tokens
Estimated from current input and output prices, before discounts.
Benchmarks
- 50%
Code Bench
Improvement over GLM-5.2 on the vendor's in-house benchmark
- 28.3
Terminal Bench 3.0
Points; GLM-5.2 scored 4.6, a gain of 23.7 points
- 66.9
DeepSWE v1.1
Points; up from 46.2 on GLM-5.2, a 20.7-point improvement
- 28.5
Agents' Last Exam CLI
Points; up from 23.8 on GLM-5.2
- 54.4%
ExploitBench
Up from 24.4% on GLM-5.2
- 84.5%
CyberGym
Vulnerability discovery benchmark
- 1769
GDPval-AA v2
Points across 44 occupational categories
Scores as published by the model provider.
Reasoning
- Default effort
- max
- Enabled by default
- Yes
- Always enabled
- Yes
- Available effort levels
- max / high / low
Quick Start
Drop-in code to call this model with Felo API Platform compatible APIs. Start in the Playground when you want to test prompts before writing code.
Get your API key
Create a key from your Felo API Platform account and set it as an environment variable.
export FELO_API_KEY="YOUR_API_KEY"Make your first request
Use this model ID with the Felo API Platform endpoint. Optional request headers such as X-Request-Id can help troubleshooting.
glm-5.3This model is reasoning-capable. If your selected protocol and model route expose reasoning controls, preserve the returned reasoning metadata when continuing a conversation.
const response = await fetch("https://openapi.felo.ai/api/v1/messages", {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.FELO_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
"model": "glm-5.3",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}),
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.json());Using third-party SDKs
Use compatible SDKs by setting baseURL to https://openapi.felo.ai/api/v1. Use the Anthropic-compatible base URL for Claude-style clients. https://openapi.felo.ai/api.
Enable streaming
Add stream: true to receive responses as server-sent events.
curl -N https://openapi.felo.ai/api/v1/messages \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
],
"stream": true
}'Endpoint
Felo API Platform exposes compatible model-calling surfaces. Pick the surface your app or agent client already supports.
Messages
Anthropic-compatible messages for Claude-style clients.
POST
https://openapi.felo.ai/api/v1/messagesParameters
modelstringglm-5.3Public model ID to call.max_tokensinteger—Maximum generated tokens.messagesarray—Conversation messages for Chat Completions.seed———stop———min_p———toolsarray—Tool calling definitions following compatible schema.top_k———top_p———logprobs———reasoning———logit_bias———temperaturenumber—Controls response variety when supported.tool_choicestring or object—Controls whether a tool should be called.top_logprobs———response_formatobject—Request structured output when supported.presence_penalty———reasoning_effort———frequency_penalty———include_reasoning———repetition_penalty———structured_outputs———parallel_tool_calls———Frequently asked questions
What is GLM 5.3?
GLM-5.3 is Z.ai's text-in, text-out model built for demanding coding and agentic workloads. It supports context of up to 1 million tokens and provides a thinking mode with three reasoning effort levels: low, high, and max. Thinking can no longer be disabled, which is a breaking change for applications that previously ran with thinking switched off. The model also offers streaming messages for real-time responses, function calling for tool integration, context caching to optimize long conversations, and structured output in formats such as JSON. It is designed for long-horizon coding agents, complex software engineering, multi-step CLI work such as refactors and CI triage, long-running agent loops, repo-scale work, and ML infrastructure diagnosis and optimization, as well as broader agentic tasks beyond coding. According to Z.ai, it is the most capable open-weights model for coding, with state-of-the-art results among open-source models on Terminal Bench 3.0 and Agents' Last Exam, and state of the art on CyberGym for vulnerability discovery. Z.ai reports that comparable work can be done at approximately 50,000 tokens instead of 120,000, and that gains concentrate on multi-step work rather than one-shot completions. GLM-5.3 uses the same base model as GLM-5.2, a 743-billion parameter Mixture-of-Experts architecture, and all improvements come from post-training. Vendor-reported figures have not yet been independently replicated, and capabilities remain significantly lower than current closed frontier models. Cyber capability is strongest in the front end of the exploitation chain, while deeper exploitation and complete offensive and defensive workflows still need improvement. Open weights were delayed for approximately two weeks after launch for a safety review.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What input and output modalities does GLM-5.3 support?
It accepts text input and produces text output.
How large is the context window?
GLM-5.3 supports context of up to 1 million tokens.
Can thinking be turned off?
No. Thinking can no longer be disabled, which is a breaking change for applications that ran with thinking switched off. The model offers low, high, and max reasoning effort levels.
What tasks is it best suited for?
It targets long-horizon coding agents, complex software engineering, multi-step CLI work, long-running agent loops, repo-scale work, ML infrastructure diagnosis and optimization, and broader agentic tasks.
How does it differ from GLM-5.2?
It uses the exact same base model as GLM-5.2, a 743-billion parameter Mixture-of-Experts architecture, and every improvement comes from post-training rather than architectural changes.
Related models
GLM 5.3 FlashX Legal
Z.aiglm-5.3-flashx-legalZ.ai's GLM 5.3 FlashX Legal is a speed-oriented, natively multimodal legal model with a context of approximately one million tokens and 128K output tokens.
Context
1.049M
Max output
131.072K
Input tokens
$0.37 / 1M tokens
Output tokens
$1.25 / 1M tokens
GLM 5.3 Flash
Z.aiglm-5.3-flashGLM 5.3 Flash is a 320B-parameter open-weight Mixture-of-Experts model from Z.ai with native text and image input, a 1 million token context window, and a design optimized for fast, inexpensive multimodal agent workloads.
Context
1.049M
Max output
943.717K
Input tokens
$0.15 / 1M tokens
Output tokens
$0.5 / 1M tokens