GLM 4.7
Z.AI's open-weight coding and reasoning model with multi-mode thinking, MIT-licensed weights, and strong agentic performance.
Overview
What is GLM 4.7
GLM 4.7 is Z.AI's open-weight language model released in December 2025, built on the GLM architecture with 358 billion parameters. It specializes in agentic coding, web search, and complex reasoning via multiple thinking modes, delivering strong scores on SWE-bench and terminal automation benchmarks.
Running it privately on Venice
On Venice, GLM 4.7 runs inside a TEE with end-to-end encryption and zero retention — your prompts are not stored or profiled. You get open-weight sovereignty with private inference, plus built-in reasoning, web search, and code-optimized output without Big-Tech surveillance.
Assessment
Strengths and limitations
- Open weights under MIT license: full sovereignty to self-host, audit, or fine-tune.
- Strong coding performance with improvements on SWE-bench and terminal tasks, plus code-optimized output on Venice.
- Multi-mode reasoning for stable complex task execution.
- Built-in web search support for grounded, autonomous workflows on Venice.
- Better UI and front-end generation than its predecessor, with cleaner layouts and modern design.
- Not uncensored: safety filters apply, which may limit certain sensitive or controversial prompts.
- Incremental upgrade over GLM-4.6 rather than a leap; for cutting-edge coding SOTA, GLM-5.1 is now Z.AI's flagship.
- Large parameter count demands significant VRAM for local self-hosting compared to smaller open models.
- Benchmark coverage is mixed: some categories remain thinly measured by independent third parties.
Capabilities
What it supports
- Tool use / function calling
- Vision (image input)
- Reasoning
- Web search
- Code-optimized
- Structured output (JSON schema)
- Audio input
- Video input
- Multiple image inputs
- Log probabilities
Specifications
Datasheet
- Maker
- Z.AI
- Released
- December 22, 2025
- Architecture
- Transformer-based GLM (General Language Model)
- Parameters
- 358 billion
- Open weights
- Yes — MIT license
- Input
- Text
- Context window
- 128K tokens
- Max output
- 32.768K tokens
- Capabilities
- Reasoning, Web search, Code-optimized
- Privacy on Venice
- Private — zero retention
- Available on Venice since
- Mar 2026
API
Call it from your code
Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "e2ee-glm-4-7-p",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Pricing
What it costs on Venice
Billed per token on Venice: $1.10 per 1M input tokens and $4.15 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Context window | Open weights | Price (Venice) | Best for |
|---|---|---|---|---|
| GLM 4.7 | 128K tokens | Yes | $1.10 in · $4.15 out / 1M | Agentic coding & reasoning |
| GLM 5.1 | 200K tokens | Yes | $1.10 in · $4.15 out / 1M | Latest GLM flagship |
| DeepSeek V3.2 | 160K tokens | Yes | $0.33 in · $0.48 out / 1M | Budget reasoning |
| Claude Sonnet 4.6 | 1M tokens | No | $3.60 in · $18 out / 1M | Closed high-capacity |
| Kimi K3 | 1M tokens | Yes | $3.75 in · $18.75 out / 1M | Long-context open |
The balanced open-weight choice for coding and reasoning with private TEE inference.
Use cases
What it is good for
- 01Agentic software development with autonomous requirement decomposition and multi-stack integration.
- 02Terminal-based automation and shell scripting in multiple languages.
- 03Conversational UI generation, webpage building, and slide design.
- 04Complex mathematical reasoning and step-by-step problem solving.
- 05Research workflows augmented with web search and structured data extraction.
Prompting
Getting better results
Break complex coding tasks into discrete steps — GLM-4.7 handles multi-file and multi-stack integration better with explicit decomposition.
Describe the full task context up front — it excels at end-to-end task completion rather than single-shot snippets.
For coding agents, specify the framework to align invocation patterns with Cline, Roo Code, or Claude Code.
Request structured JSON output explicitly when integrating the model into external pipelines.
Version history
Predecessor with lighter coding and reasoning capabilities.
Current — improved agentic coding, web search, and multi-mode reasoning.
FAQ
Frequently asked questions
GLM 4.7 is Z.AI's open-weight language model released in December 2025. It is optimized for agentic coding, complex reasoning, and web search, and is licensed under MIT for self-hosting or fine-tuning.
Venice bills GLM 4.7 at $1.10 per 1 million input tokens and $4.15 per 1 million output tokens, with no subscription required.
Yes. GLM 4.7 ships with open weights under the MIT license, so you can self-host, audit, or fine-tune it. The weights are available on Hugging Face.
You can try it on Venice with free-tier credits; ongoing use is billed per token. Self-hosting the open weights is free of licensing fees, though you pay for your own compute.
GLM 4.7 is a proven, cost-efficient workhorse for coding and reasoning. GLM 5.1 offers a larger 200K context and stronger SOTA performance on complex software engineering tasks. Many teams route simple tasks to 4.7 and hard tasks to 5.1.
Yes. On Venice, GLM 4.7 supports web search alongside reasoning and code-optimized output, enabling grounded research and complex coding workflows without leaving the chat.
No. GLM 4.7 is not uncensored and includes safety filters. For fully uncensored open-weight inference, you would need to use a different model or fine-tune the weights yourself.
On Venice, GLM 4.7 runs in a TEE with end-to-end encryption and zero retention — your prompts are not stored, profiled, or used for training. Alternatively, download the MIT-licensed weights and self-host for complete data sovereignty.
Run GLM 4.7 privately
No prompt logging. No data used for training.