Close Menu
  • AI
  • Content Creation
  • Tech
  • Robotics
AI-trends.todayAI-trends.today
  • AI
  • Content Creation
  • Tech
  • Robotics
Trending
  • A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels
  • Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI
  • Dragos Finds Immediate Leaks and Stolen Keys in AI Provide Chains
  • Lean and Mathlib Communities Certify Claude-Assisted Riemann Proof – Unite.AI
  • AWS Routes HyperPod Inference Throughout Areas From One Endpoint – Unite.AI
  • DailyAI Briefing: A Lean-checked Riemann proof, plus 4 extra
  • Anthropic Shares Drop 8.5% in Second Session Throughout AI Selloff – Unite.AI
  • OpenAI Ships GPT-Stay-1.1 for Voice Brokers, Cuts Idle Value to $0.02/Min
AI-trends.todayAI-trends.today
Home»Tech»Zhipu AI releases GLM-4.7.Flash, a 30B-A3B MOE model for efficient local coding and agents

Zhipu AI releases GLM-4.7.Flash, a 30B-A3B MOE model for efficient local coding and agents

Tech By Gavin Wallace20/01/20264 Mins Read
Facebook Twitter LinkedIn Email
NVIDIA Releases Llama Nemotron Nano 4B: An Efficient Open Reasoning
NVIDIA Releases Llama Nemotron Nano 4B: An Efficient Open Reasoning
Share
Facebook Twitter LinkedIn Email

GLM-4.7 Flash is a member of GLM 4.7 and is designed for developers looking to get strong reasoning and coding performance from a model which can be run locally. Zhipu AI’s (Z.ai), describes GLM-4.7 as 30B A3B MoE and calls it the most powerful model of the 30B family, designed to be lightweight and efficient.

GLM 4.7 model class and position within the family

GLM4.7-Flash has 31B parameter types and includes BF16, F32, and architecture tags. glm4_moe_lite. The GLM-4.7-Flash is a conversational device that supports both English and Chinese. The GLM 4.7-Flash is part of the GLM 4.7 collection, alongside the GLM 4.7 and GLM 4.7-FP8 models.

Z.ai offers GLM-4.7 Flash as a lightweight and free tier deployment option compared to GLM-4.7, but still targets coding, reasoning and text generation. It is therefore attractive to developers that cannot use a model of the 358B-class but want an MoE with strong benchmarks and a modern design.

The length of the architecture and context

The model in a Mixture of Experts Architecture of this kind stores more parameters for each token than the number of them it actually activates. It allows for specialization between experts, while maintaining the compute efficiency per token at a level closer to that of a small dense model.

GLM Flash version 4.7 is compatible with context sizes up to 128k and has a high performance when coding against models that are similar in size. This size context is ideal for codebases with multiple files, long technical documents and multi-file repositories. Many existing models require aggressive chunking.

GLM 4.7-Flash is a flash version of the standard language model interface. It also includes a chat tool, which makes it easy to integrate into LLM systems with little change.

Benchmark Performance in Class 30B

The Z.ai team compares GLM-4.7-Flash with Qwen3-30B-A3B-Thinking-2507 and GPT-OSS-20B. The GLM-4.7 Flash is the leader or competitive in a variety of benchmarks including math, reasoning and long-horizon.

https://huggingface.co/zai-org/GLM-4.7-Flash

Tables like the one above show that GLM-4.7 Flash is among the top models of its 30B class. GLM 4.7-Flash, not only is it a compact GLM deployment but is also an extremely high-performing model when compared to established benchmarks in coding and agents.

Thinking and evaluating parameters

Temperature 1.0, Top p 0,95, and Max new tokens are the default settings for most tasks. The sampling is relatively wide with a high budget.

In the case of Terminal Bench, SWE Bench Verified or both, temperature is set to 0.7 degrees Celsius, top P at 1.0, with a maximum number of new tokens of 16384. For τ²-Bench, the configuration uses temperature 0 and max new tokens 16,384. This stricter setting reduces randomness in tasks that require stable tool usage and multiple step interactions.

Z.ai team also recommends turning on Preserved Thinking mode for multi turn agentic tasks such as τ²-Bench and Terminal Bench 2. This mode keeps internal reasoning tracks across turns. This is especially useful for agents with long function chains and corrections.

GLM-4.7 Flash fits well with developer workflows

The Flash version of GLM-4.7 combines multiple properties for agentic applications that require coding.

  • The 30B-A3B MoE has 31B parameters and 128k context tokens.
  • Strong benchmark results on AIME 25, GPQA, SWE-bench Verified, τ²-Bench, and BrowseComp compared to other models in the same table.
  • There are also documented evaluation criteria and a preserved thinking mode that can be used for tasks involving multiple turns.
  • First-class support for SGLang, vLLM and Transformers inference based on ready to use commands.
  • Hugging Face’s ecosystem is undergoing a series of refinements and quantizations including MLX converters.

Click here to find out more Model weight. Also, feel free to follow us on Twitter Don’t forget about our 100k+ ML SubReddit Subscribe now our Newsletter. Wait! Are you using Telegram? now you can join us on telegram as well.


Michal is a professional in the field of data science with a Masters of Science degree from University of Padova. Michal Sutter excels in transforming large datasets to actionable insight. He has a strong foundation in statistics, machine learning and data engineering.

AI coding
Share. Facebook Twitter LinkedIn Email
Avatar
Gavin Wallace

Related Posts

A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels

22/11/2026

OpenAI Ships GPT-Stay-1.1 for Voice Brokers, Cuts Idle Value to $0.02/Min

21/11/2026

Meet Pondering Machines Inkling 2 Mini: 26B Looped MoE Runs on a 32GB Mac

21/11/2026

Baseten Releases Bison-1 and Groundhog-1, Topping Choice Mannequin Index

21/11/2026
Top News

Jensen Huang’s Imaginative and prescient for Europe Runs on Plug-In Information Facilities

Google, Bing and Private Claude Chats: Exposed!

AI Models Can Also Get Brain Rot

Moltbook – the social network for AI Agents – exposed data of real humans

It’s not for plumbers or electricians that the real AI talent war is.

Load More
AI-Trends.Today

Your daily source of AI news and trends. Stay up to date with everything AI and automation!

X (Twitter) Instagram
Top Insights

Authors Are Posting TikToks to Protest AI Use in Writing—and to Prove They Aren’t Doing It

18/06/2025

Andrew Ng Has Released OpenWorker – An Open Source, Local First Desktop AI Assistant That Delivers Finished Deliverables instead of Chat

23/07/2026
Latest News

A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels

22/11/2026

Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI

22/11/2026
X (Twitter) Instagram
  • Privacy Policy
  • Contact Us
  • Terms and Conditions
© 2026 AI-Trends.Today

Type above and press Enter to search. Press Esc to cancel.