GLM-5.2 Officially Released — 753B Parameters, 1M Context, MIT License

Z.ai (formerly Zhipu AI) has released its latest flagship model, **GLM-5.2**, on Hugging Face. It has 753B parameters, supports a context length of **1M tokens**, and is available as fully open source under the MIT license.

The benchmark results are what make it stand out. GLM-5.2 is **the first open-weight model to break 80% on Terminal-Bench** (81.0%) and took second place in WebDew Arena. In coding, it scores 62.1% on SWE-bench Pro and 74.4% on FrontierSWE (Dominance) — a major leap from the previous-generation GLM-5.1.

On the architecture side, it introduces a technique called "IndexShare," which cuts per-token FLOPs at 1M context by 2.9×. It also refines the MTP layers, boosting the acceptance length for speculative decoding by up to 20%.

Community reaction has been lively: on Reddit's LocalLLaMA, the verdict is that "GLM-5.2 outperforms every available open model." It is already available on HuggingChat.

The open-weight release had been teased in a report on June 13, and this official launch is what reveals the full benchmark results and technical specifications.

ChatGPT's Market Share Falls Below 50% for the First Time

According to TechCrunch, ChatGPT's **share of the AI assistant market has dropped below 50% for the first time**.

That said, ChatGPT's own user base is not shrinking. It exceeds **1.1 billion** monthly active users and remains the world's largest AI assistant. The share decline comes down to competitors simply growing faster.

In second place is Google's **Gemini** with 662 million monthly active users, and third is Anthropic's **Claude** with 245 million. Gemini's climb is especially pronounced, likely driven by pre-installation on Android devices and tight integration with the Google ecosystem.

As the AI assistant market as a whole keeps expanding, this stands out as a symbolic data point showing that ChatGPT's single-player dominance is gradually starting to shift.

Android 17 Released — Expanded Gemini Features

Google has released **Android 17** and **Wear OS 7**. The release includes multitasking features, parental controls, security tools, and smartwatch enhancements.

On the AI front, Google's latest AI models are being delivered to Pixel devices through Pixel Drop. However, according to Bloomberg, some flagship AI features are expected to arrive "in the coming months," and it appears a few of them will not be part of the initial release.

Android 17 rolls out to Pixel devices first, with other manufacturers' devices to follow.

Mistral Teases a New Model Family for July

Word from the Reddit LocalLLaMA community is that Mistral AI plans to **release a new open-weight model family in July**. Detailed specs and licensing terms are unknown at this point, but with GLM-5.2 heating up the open-source LLM race, all eyes are on Mistral's next move.

AWS Brings P-EAGLE Speculative Decoding to SageMaker AI

On the AWS Machine Learning Blog, AWS announced that inference acceleration using **P-EAGLE (Parallel-EAGLE)** is now available on SageMaker AI.

P-EAGLE evolves EAGLE's draft token generation from sequential processing to **fully parallelized** execution. It delivers up to 1.69× higher throughput compared to EAGLE-3, and up to roughly 4× faster inference versus the baseline (standard inference).

On SageMaker JumpStart, P-EAGLE comes pre-configured for four models — GPT-OSS-120B, GPT-OSS-20B, Qwen3-Coder-30B-A3B-Instruct, and Gemma-4-31B-IT — deployable in one click.

---