Models / ZAI

Reasoning

Chat

Code

GLM-5.2

Flagship open reasoning model for agentic software engineering with 256K context window

Try now

read docs

About model

GLM-5.2 is Z.ai's newest flagship model for agentic coding and long-horizon software engineering, built on the GLM-5 Mixture-of-Experts backbone — 744B total parameters with 40B active per token. The release also adds two thinking-effort levels, letting developers trade compute depth against throughput depending on task complexity. GLM-5.2 ships under the MIT License with open weights. According to Z.ai, the model delivers powerful coding capabilities and continued strengths in long-horizon agentic tasks.

Token Context Window

256K

Usable 256K-token context window with a 131,072-token output cap

Total Parameters (40B Activated)

744B

Mixture-of-Experts backbone optimized for coding agents

Thinking Effort

Configurable

Two thinking-effort levels exposed directly at the API level

Model key capabilities

Long-Context Agentic Coding: A usable 256K-token context window — 5× larger than GLM-5.1's ~200K limit — with a 131,072-token output cap. According to Z.ai, the model tracks cross-file dependencies across full repositories in a single pass, enabling whole-repo refactors without repeated context retrieval.
Configurable Thinking Effort: Two thinking-effort levels exposed at the API level let developers dial between lighter inference for fast tasks and deeper reasoning for complex long-horizon sessions — a tradeoff between throughput and reasoning depth.
Drop-in Agent Compatibility: Day-one support for eight coding agents including Claude Code, Cline, Roo Code, Goose, and OpenCode via an OpenAI-compatible and Anthropic Messages API — a config-level swap with no framework changes required.

Quickstart guides

Agents

Build a Coding Agent (Webinar)

Performance benchmarks

Model	FrontierMath Tier 4	GPQA Diamond	HLE	SciCode	GDPval-AA	Terminal-Bench 2.1	Agent Arena	FrontierCode	DeepSWE
GLM-5.2	29.3%	89.5%	40%	50%	50%	78%	+9.1pp	24.5%	44%
Related open-source models
Competitor closed-source models
Claude Fable 5	87.8%	92.6%	53%	60%	62%	85%		53.5%	70%
Claude Opus 5	73.2%	93.2%	53%	56%	68%	89%	+16.4pp	53.4%	74%
GPT-5.6 Sol	82.9%	94.1%	47%	56%	61%	88%		47.5%	73%
Grok 4.5	24.4%	93.1%	40%	54%	51%	82%	+4.2pp	42.4%	54%
GPT-5.6 Luna	61.0%	91.1%	37%	53%	54%	81%		39.8%	67%

API usage

cURL
Python
Typescript

Endpoint:

zai-org/GLM-5.2

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-5.2",
    "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-5.2",
  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-5.2',
  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:
• 744B total parameter MoE architecture with 40B parameters activated per token
• 256K context window with 131,072 max output tokens
• Supports configurable thinking modes (two effort levels) for step-by-step reasoning
• Open-weights distribution under the MIT License

Training Methodology:
• Pre-trained and instruction-tuned entirely on Huawei Ascend 910B hardware under the MindSpore framework
• Post-training optimizations targeting repo-scale dependencies and extended plan-then-execute loops

Performance Characteristics:
• Flagship reasoning performance built specifically for long-horizon orchestration
• Native framework compatibility with multiple coding agent workflows
• No official benchmarks were published at launch; performance metrics will be updated as they are verified
‍
Prompting
Together AI API Access:
• Access GLM-5.2 via Together AI APIs using the endpoint zai-org/GLM-5.2
• Authenticate using your Together AI API key in request headers
• Supports configurable thinking effort parameters
• Available on Together AI serverless infrastructure
‍
Applications & use cases
Repository-Scale Engineering:
• Complete multi-file refactoring, code migration, and complex system updates in a single pass
• Navigation of large codebases without needing vector-search retrieval chunking

Autonomous Technical Workflows:
• Extended plan-then-execute cycles utilizing up to 131K output tokens for deep tracing
• Multi-step tool use, system debugging, and algorithmic optimization workflows

Drop-in Agent Integrations:
• Full compatibility with Claude Code, Cline, Roo Code, Goose, OpenCode, and custom frameworks
• Switch configurations smoothly via OpenAI-compatible and Anthropic Messages API formats
‍

Related models

Model specifications

Model data

Model provider
ZAI
Type
Reasoning
Chat
Code
Main use cases
Reasoning
Features
Function Calling
JSON Mode
Deployment
Serverless
Provisioned Throughput
Endpoint
zai-org/GLM-5.2
Parameters
753B
Activated parameters
40B
Context length
256K
Input price
$1.40 / 1M tokens
$0.26 (cached)/1M
Output price
$4.40 / 1M tokens
Input modalities
Text
Output modalities
Text

Released
June 16, 2026
Quantization level
FP4
Category
Chat

Run in Playground

Quickstart docs

Deploy model

GLM-5.2

About model

API usage

Model card

Prompting

Applications & use cases