Agentic software development
Plan and execute multi-file coding work with terminal tools, repository context, and iterative problem solving.
Independent tool overview
GLM-4.7 at a glance
GLM-4.7 is Z.AI's MIT-licensed 358B mixture-of-experts language model for agentic coding, tool use, reasoning, and bilingual English-Chinese work, available through an API or self-hosting.
Visit the official GLM-4.7 site ↗
- Released
- December 22, 2025
- Best for
- Agentic coding, tool use, and bilingual reasoning
- Context window
- 200K tokens
- Maximum output
- 128K tokens
- Model size
- About 358B total parameters, mixture of experts
- License
- MIT
- API price
- $0.60 input / $2.20 output per 1M tokens
Overview
What GLM-4.7 is
GLM-4.7 is an open-weight text model Z.AI released in December 2025 with a strong emphasis on completing multi-step coding and agent tasks. It can reason before acting, call tools, produce structured output, stream responses, and work with a 200K-token context window. The full checkpoint contains about 358 billion parameters and uses a mixture-of-experts architecture, so local deployment is aimed at serious inference infrastructure rather than an ordinary laptop.
Developers can call GLM-4.7 through Z.AI's OpenAI-compatible API, connect it to coding agents such as Claude Code, Cline, Kilo Code, and Roo Code, or run the MIT-licensed weights with supported engines including vLLM and SGLang. The family also includes the lower-cost GLM-4.7-FlashX and free GLM-4.7-Flash models.
GLM-4.7 remains available, but it is no longer Z.AI's newest flagship. GLM-5 and later GLM releases have moved the product line forward, so new projects should benchmark GLM-4.7's lower API price and mature open checkpoint against the latest models before standardizing on it.
Use cases
Who GLM-4.7 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Cost-conscious API workloads
Use a capable reasoning and coding model at lower token prices than many frontier proprietary APIs.
Self-hosted model programs
Deploy MIT-licensed weights in controlled infrastructure when data location, customization, or provider independence matters.
English-Chinese applications
Build coding, writing, translation, long-context, and agent workflows that operate across both supported languages.
Capabilities
Core GLM-4.7 features
Agentic coding
Focuses on task completion, including requirement interpretation, decomposition, multi-technology integration, terminal work, and repository-scale edits.
Thinking modes
Supports reasoning controls, including preserved thinking for multi-turn agent tasks where intermediate reasoning continuity matters.
Tool and function calling
Can invoke external tools and APIs for browsing, coding agents, and other multi-step workflows.
200K context
Handles long codebases and documents with up to 200K tokens of context and up to 128K output tokens.
OpenAI-compatible API
Works with Z.AI's chat-completions endpoint and common OpenAI SDK patterns, reducing integration work.
Open weights
The 358B-parameter checkpoint is published under the MIT license and supports self-hosting with engines such as vLLM and SGLang.
Flash variants
GLM-4.7-FlashX offers much lower token prices, while GLM-4.7-Flash is listed as free for lightweight workloads.
Process
How the GLM-4.7 workflow works
Step 1
Choose a deployment path
Use the Z.AI API for managed inference, a coding-plan endpoint for supported IDE agents, or self-host the open weights.
Step 2
Select the family member
Compare full GLM-4.7 for quality, FlashX for inexpensive throughput, and Flash for free lightweight access.
Step 3
Configure reasoning and tools
Set thinking behavior, streaming, structured output, function definitions, and context limits for the workload.
Step 4
Test on representative tasks
Measure repository edits, tool-call reliability, latency, token consumption, and regressions on your own evaluations.
Step 5
Monitor and upgrade deliberately
Track price and quality against newer GLM releases, then pin model identifiers and test before changing production versions.
Cost
GLM-4.7 pricing and free plan
Z.AI charges the managed API per token and separately offers coding subscriptions. GLM-4.7 costs $0.60 per million input tokens, $0.11 per million cached input tokens, and $2.20 per million output tokens. FlashX is cheaper and Flash is listed as free. Self-hosting has no model license fee under MIT, but the full 358B checkpoint has substantial infrastructure costs.
GLM-4.7 API
$0.60 input / $2.20 output per 1M tokens
Managed full-model inference through Z.AI's general API.
- $0.11 per 1M cached input tokens
- Cached-input storage listed as free for a limited time
- OpenAI-compatible endpoint
GLM-4.7-FlashX API
$0.07 input / $0.40 output per 1M tokens
Lower-cost, higher-speed member of the GLM-4.7 family.
- $0.01 per 1M cached input tokens
- 200K context
- Text input and output
GLM-4.7-Flash API
Free
The lightweight free-tier model in the family.
- Input, cached input, and output listed as free
- Usage limits and availability may apply
- 200K context
Self-hosted GLM-4.7
No model license fee
Download and operate the MIT-licensed model on your own infrastructure.
- About 358B total parameters
- Compute, memory, storage, and operations are your responsibility
- vLLM and SGLang are documented inference paths
GLM Coding Plan Lite
$18/month
Subscription access for supported coding agents, with annual billing shown at an effective $12.60 per month.
- Designed for one lightweight project at a time
- Supports 20+ coding tools
- Access rotates toward newer flagship models
- Coding-plan quota cannot be used as a general application API
Pricing checked . Check current pricing at the source ↗
Assessment
GLM-4.7 strengths and limitations
Where it stands out
- Strong emphasis on complete agentic coding tasks rather than isolated code snippets.
- Low managed-API prices relative to many closed frontier coding models.
- MIT-licensed weights provide real self-hosting and customization flexibility.
- Large 200K context window and unusually high 128K maximum output support long workflows.
- OpenAI-compatible API and integrations with popular coding agents simplify adoption.
- Flash and FlashX variants let teams trade capability for price and speed within the same family.
What to consider
- GLM-4.7 has been superseded by newer GLM-5-series models, so it is not the default choice for maximum current capability.
- The full 358B checkpoint requires substantial GPU memory, storage, and inference expertise to self-host well.
- Z.AI's published benchmark results are vendor-reported and should be validated on your own repositories and agent tasks.
- The model accepts and produces text; it is not the right GLM variant for native image, audio, or video understanding.
- Very long context and output limits do not guarantee reliable recall or completion across the entire window.
- Coding-plan subscriptions and general API billing are separate products with different usage rules.
Compare
GLM-4.7 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
GLM 5.2
A newer Z.AI flagship for longer-horizon coding work and substantially larger usable context.
Explore GLM 5.2 →Coding
Qwen3-Coder-Next
A smaller open-weight agentic coding model that can be easier and cheaper to self-host.
Explore Qwen3-Coder-Next →Business Operations
DeepSeek
Another cost-focused model family with open releases and strong reasoning and coding options.
Explore DeepSeek →Coding
Claude Code
A polished coding-agent product for teams that prioritize the complete development workflow over operating a model directly.
Explore Claude Code →Questions
GLM-4.7 FAQs
Is GLM-4.7 open source?
Its model weights are publicly available on Hugging Face under the MIT license. That permits broad commercial use and modification, but operating the full checkpoint still requires substantial infrastructure.
How large is GLM-4.7?
The official Hugging Face checkpoint is about 358 billion total parameters and uses a mixture-of-experts architecture. Independent vLLM deployment documentation describes roughly 32 billion active parameters per token.
How much does the GLM-4.7 API cost?
Z.AI lists $0.60 per million input tokens, $0.11 per million cached input tokens, and $2.20 per million output tokens. GLM-4.7-FlashX is cheaper and GLM-4.7-Flash is listed as free.
What is GLM-4.7 best at?
Its main strengths are agentic coding, tool use, multi-step reasoning, frontend generation, long-context text work, and English-Chinese applications.
Can GLM-4.7 work with Claude Code?
Yes. Z.AI documents GLM-4.7 for Claude Code and other coding tools through its OpenAI- or coding-plan-compatible endpoints. It is a Z.AI model, not an Anthropic model.
Should I use GLM-4.7 or GLM-5?
Use GLM-4.7 when its lower price, stable integration, or open checkpoint meets your needs. Test newer GLM-5-series models when you need stronger current reasoning, coding, or long-horizon agent performance.
Does GLM-4.7 support images?
No. The documented GLM-4.7 input and output modality is text. Z.AI maintains separate vision-language and image-generation models.
Bottom line
Our GLM-4.7 verdict
GLM-4.7 remains an attractive value model for agentic coding: it is inexpensive through Z.AI's API, has a generous context and output envelope, integrates with common coding agents, and offers MIT-licensed weights. Its biggest tradeoff is timing—newer GLM generations now lead the family—so teams should choose it for cost, deployment control, or proven workload fit rather than assume it is still Z.AI's highest-capability model.
Visit GLM-4.7 website ↗Similar Coding Tools
View all Coding AI tools →
MiniMax-M2.1
MiniMax 2.1 - New model with powerful capabilities across a variety of programming languages and for mobile and web app development
NousCoder-14B
NousCoder-14B - Nous Research's competitive olympiad programming model
GPT-5.2-Codex
GPT-5.2-Codex - Agentic coding model for professional software engineering and defensive cybersecurity.
SERA
SERA - AI2's open-source coding agents able to be cheaply trained on private codebases with native support for Claude Code.
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.