GLM 4.7 Flash
Z.ai's lightweight 30B MoE text model built for fast coding, tool use, and agentic tasks.
Overview
What is GLM 4.7 Flash
GLM 4.7 Flash is a lightweight, open-weights text model from Z.ai. Released in January 2026, it is a 30B-parameter mixture-of-experts model with 3B active parameters per token, built for fast coding, tool use, and agentic tasks. It runs under an MIT license and offers strong benchmark performance for its size class.
Running it privately on Venice
On Venice, GLM 4.7 Flash runs under a private, zero-retention tier — your prompts are not stored or used for training. You get the same open-weights model with tool use, reasoning, and web search capabilities, served with fp8 quantization for efficient inference without Big-Tech surveillance.
Assessment
Strengths and limitations
- Strong coding and agentic performance for its size — scores 59.2 on SWE-bench Verified and 91.6 on AIME 2025 per the official model card.
- Supports tool use, reasoning, and web search, plus structured JSON output for building automated workflows.
- Open-weights MIT license lets you inspect, modify, or self-host the model outside of Venice.
- Efficient fp8 quantization on Venice keeps inference fast and affordable without sacrificing core capabilities.
- Offers thinking modes for complex multi-step tasks, improving reliability on debugging and agentic execution.
- Text-only: no vision or image input support.
- Not fully uncensored: safety alignment is present, so it may refuse certain sensitive or restricted prompts.
- Smaller active parameter count (3B per token) can limit depth on the most complex reasoning tasks compared to full-scale frontier models.
- Venice hosts the fp8 quantized version; self-hosters seeking maximum fidelity may want to run the original full-precision weights locally.
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
- January 19, 2026
- Architecture
- Mixture-of-Experts (MoE)
- Parameters
- 30B total (3B active per token)
- Modality
- Text → Text
- Open weights
- Yes — MIT license
- Context window
- 128K tokens
- Max output
- 16.384K tokens
- Capabilities
- Function calling, Reasoning, Web search
- Privacy on Venice
- Private — zero retention
- Available on Venice since
- Jan 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": "zai-org-glm-4.7-flash",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Pricing
What it costs on Venice
Billed per token on Venice: $0.13 per 1M input tokens and $0.50 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 | Best for | Context window | Open weights | Price (Venice) |
|---|---|---|---|---|
| GLM 4.7 Flash | Lightweight MoE with tool use, reasoning, and web search. Best price-to-performance for high-volume coding. | 128K tokens | Yes | $0.13 in · $0.50 out / 1M |
| DeepSeek V3.2 | Larger-context open rival with similar output pricing and strong generalist performance. | 160K tokens | Yes | $0.33 in · $0.48 out / 1M |
| Google Gemma 4 31B Instruct | Google's 31B open model with a very wide context window and slightly lower per-token cost. | 256K tokens | Yes | $0.12 in · $0.36 out / 1M |
| Kimi K2.6 | Open-weights long-context model with strong reasoning, but significantly more expensive per token. | 256K tokens | Yes | $0.75 in · $3.50 out / 1M |
Lightweight MoE with tool use, reasoning, and web search. Best price-to-performance for high-volume coding.
Use cases
What it is good for
- 01High-throughput coding assistants and autocomplete where low latency matters.
- 02Agentic workflows that need function calling, web search, and structured JSON output.
- 03Lightweight self-hosted deployment for privacy-sensitive teams (open weights, MIT license).
- 04Multi-turn debugging and reasoning tasks using the model’s thinking mode.
- 05Cost-effective text generation for startups and indie developers.
Prompting
Getting better results
Enable thinking mode for complex debugging or multi-step agent tasks — it improves accuracy at the cost of slightly higher latency.
Use structured output (JSON schema) when integrating with tools or APIs to get predictable, parseable responses.
For coding, specify the language and framework explicitly; the model is tuned for precise code generation.
Keep context within Venice’s 128K token window for best results, even though the native model supports 200K.
Version history
Base model in the GLM-4.7 series.
Current lightweight, speed-optimized variant.
FAQ
Frequently asked questions
GLM 4.7 Flash is a lightweight, open-weights text model from Z.ai (Zhipu AI). Released in January 2026, it is a 30B-parameter MoE model with 3B active parameters per token, designed for fast coding, tool use, reasoning, and web search.
Venice bills per token at $0.13 per 1M input tokens and $0.50 per 1M output tokens. There is no subscription required; you pay only for what you use.
Yes. The model weights are released under an MIT license and available on Hugging Face, so you can self-host or modify it. On Venice, it runs privately with zero retention.
Yes. On Venice it supports function calling, reasoning, web search, and structured JSON output, making it suitable for agentic workflows and automated pipelines.
GLM 4.7 Flash is cheaper on input and offers strong coding benchmarks for its size, while DeepSeek V3.2 provides a larger 160K context window and very similar output pricing. Choose Flash for cost efficiency; choose DeepSeek V3.2 if you need the extra context.
No. While it is open-weights and private on Venice, it is not marketed as uncensored and retains standard safety alignment. It may refuse certain sensitive requests.
Venice hosts GLM 4.7 Flash with a 128K token context window and up to 16,384 tokens max output per response.
Yes. The model supports thinking modes for multi-step reasoning and complex agentic tasks, which improves accuracy on debugging and advanced coding problems.
Run GLM 4.7 Flash privately
No prompt logging. No data used for training.