The first open-weight model that feels right in coding harnesses as a general agent. Released June 13, 2026 — it's already mixing it up with OpenAI and Anthropic on the Arena agent leaderboard.
GLM-5.2 was rolled out unusually on a Saturday, June 13th, 2026, to GLM Coding Plan members, with the official MIT-licensed weights and release blog following three days later on June 16th. On paper, a minor version bump from the popular GLM-5.1. In practice, it crossed a meaningful user-experience threshold — becoming the first open model you can drop into a coding harness and genuinely trust as a general agent.
The first open-weight model that works right inside coding harnesses like Claude Code — a credible alternative to frontier closed models for real agentic coding work.
Trained with Z.ai's popular SLIME RL framework. The community recommendation is clear: always run GLM-5.2 on Max thinking effort for best results.
Fully MIT-licensed, downloadable weights on HuggingFace. Run it on Fireworks, Together, Thinky (via Tinker), Prime Intellect, or your own infra. No silent safeguards, no gated access.
On Design Arena — a benchmark with real teeth among actual designers — GLM-5.2 bested Claude Fable itself, the recently-banned frontier hype machine.
The only open model mixing it up with OpenAI and Anthropic's latest — matching Claude Opus 4.8's no-thinking effort using its Max mode.
The community focal-point of discussion has only been this clear for an open release once before — DeepSeek R1. GLM-5.2 has well exceeded that.
What made Kimi K2 impressive was that big steps in open performance could come from anywhere in China. The step GLM-5.2 has taken is more of a one-way door for AI progress. As AI systems get more complex and expensive to build — with tools, integrated harnesses, and scaled weights — it was not a given that this moment would happen at all. Anthropic's record revenue growth on the back of Claude Code is driven by being the best model and the only model that can really do this. GLM-5.2 is the first of many coming open-weight models to offer credible alternatives.
Community benchmarks from the June 16th release showed consistently better-than-expected results. Illustrative standings based on the article's reported leaderboard placements.
Pretty much everyone in the AI commentariat and researcher class praised the model after using it personally.
"Genuinely impressed, almost shocked, at how good GLM-5.2 by @zai_org is at coding. This changes things."
— CEO of Vercel"Open-weight Fable capabilities will be here sooner than Q1 2027."
— Z.ai Founder, to Elon Musk"Such a focal point of discussion among the community has only been so clear with an open model release once before — DeepSeek R1. GLM-5.2 has well exceeded that."
— Interconnects AIGLM-5.2 is available across Fireworks, Together, Thinky (Tinker), Prime Intellect, and self-hosted. Use Max thinking effort.
# Install your preferred provider SDK
from fireworks.client import Fireworks
client = Fireworks()
response = client.chat.completions.create(
model="zai/glm-5.2",
max_tokens=8192,
thinking_effort="max", # ← always recommended
messages=[
{"role": "system",
"content": "You are an expert coding agent."},
{"role": "user",
"content": "Refactor this repo into a CLI tool with tests."},
],
)
print(response.choices[0].message.content)The first open-weight model that genuinely belongs in your coding agent loop.