# GLM-4.7

> **TL;DR.** GLM-4.

- **Category:** AI / Models / Open-Weight
- **Stage:** rising
- **Age:** 176 days
- **Origin date:** 2025-12-22
- **First detected:** 2026-04-30
- **Canonical URL:** https://earlyterms.com/term/glm-4-7
- **Sources:** 7 primary URLs

## Definition

GLM-4.7 is an open-weight large language model from Z.ai (formerly Zhipu AI) designed for agentic coding, multi-step reasoning, and production tool-use workflows. The 358B-parameter Mixture-of-Experts model activates 32B parameters per inference, runs under an MIT license, and supports a 202k-token context window.

Z.ai [released GLM-4.7 on December 22, 2025](https://www.prnewswire.com/news-releases/zai-releases-glm-4-7-designed-for-real-world-development-environments-cementing-itself-as-chinas-openai-302649821.html), with the HN launch thread accumulating 437 points and 235 comments within hours. The model introduces three thinking modes — Interleaved, Preserved, and Turn-level Thinking — that keep reasoning intact across multi-turn conversations, a gap that hurt earlier GLM versions on long agentic tasks.

## Analogy

Think of it as a diesel-engine workhorse: fewer cylinders firing at once, but built to run all day under load.

## Why it's emerging now

GLM-4.7 arrived December 22, 2025 as the first open-weight model to match Claude Sonnet 4.5 on SWE-bench Verified (73.8%) while costing $0.38 per million input tokens via API. Z.ai's Tsinghua-rooted lab positioned it specifically for Claude Code and Cline workflows — open-source agentic coding at a price that undercuts every Western frontier model.

## Related terms

- *child:* GLM-5.1
- *child:* GLM-4.7-Flash
- *competitor:* Qwen3
- *competitor:* DeepSeek V4
- *parent:* agentic-coding
- *competitor:* Claude Opus 4.7
- *competitor:* Kimi K2.6
- *related:* coding-agents
- *related:* claude-code
- *related:* roo-code

## Sources

1. [Z.ai press release — GLM-4.7 launch (Dec 22, 2025)](https://www.prnewswire.com/news-releases/zai-releases-glm-4-7-designed-for-real-world-development-environments-cementing-itself-as-chinas-openai-302649821.html)
2. [HuggingFace model card — zai-org/GLM-4.7 (architecture, benchmarks, usage)](https://huggingface.co/zai-org/GLM-4.7)
3. [HN — 'GLM-4.7: Advancing the Coding Capability' (437 pts, 235 comments)](https://news.ycombinator.com/item?id=46357287)
4. [Cerebras — GLM-4.7 at 1,000 tokens/sec on Cerebras Inference Cloud](https://www.cerebras.ai/blog/glm-4-7)
5. [OpenRouter — GLM-4.7 API pricing ($0.38/M input, $1.74/M output)](https://openrouter.ai/z-ai/glm-4.7)
6. [Techloy — '$200 vs $3' framing of Claude Max vs GLM-4.7 coding plan](https://www.techloy.com/americas-200-ai-coding-tool-just-met-a-3-chinese-rival-glm-4-7/)
7. [HN — GLM-4.7-Flash launch (378 pts, 135 comments, Jan 19, 2026)](https://news.ycombinator.com/item?id=46679872)

---
_Generated by EarlyTerms · https://earlyterms.com/term/glm-4-7_