May 26 AI Briefing · Issue #327

2026-05-26 00:00

Author: RadarAI Editorial Editor: RadarAI Editorial Last updated: 2026-05-26 Review status: Editorial review pending Brief 速报官方 AI动态开源

Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.

## 🔍 Key Insights **DeepSeek** launches a low-price offensive with its permanently discounted V4-Pro API, targeting the **Claude Code**–level programming agent market; **Baishan Intelligence** breaks the edge-side bottleneck by achieving **1.58-bit ternary quantization** training of a **60-billion-parameter model** on Huawei's Ascend platform—reducing GPU memory usage by 6× while retaining 97% of model capability [2]; meanwhile, **Gemini** quietly overhauls its billing logic, drastically shrinking actual usage quotas for paying users and exposing a trust gap in large-model commercialization [10]. ## 🚀 Major Updates - **DeepSeek adopts the 'Mixue Ice Cream & Tea' strategy to build China's answer to Claude Code** [0]: The V4-Pro model API is permanently discounted, while DeepSeek simultaneously advances R&D of an AI programming agent tool comparable to Claude Code. - **The bottleneck preventing 60-billion-parameter LLMs from running on smartphones has finally been cracked—by a Chinese AI company** [2]: Baishan Intelligence, in collaboration with Tsinghua University and OpenBMB, releases BitCPM-CANN—a ternary-quantized model trained end-to-end on Huawei's Ascend platform. - **Gemini has 'defected,' turning paying users into the punchline** [10]: Google shifts Gemini's billing model from per-call pricing to compute-consumption-based pricing—and introduces dual usage caps—sparking widespread user backlash. - **RenderFlow: Baidu's agentic code delivery in vertical search services** [9]: Baidu deploys an LLM-powered agentic system, compressing code delivery cycles for individual use cases from days down to minutes. - **Ant Group leads investment; Guanglun Intelligence secures $2B valuation, leading global embodied AI data infrastructure** [6]: The value center of the embodied intelligence industry is rapidly shifting toward **data and evaluation infrastructure**. - **Breaking! 'AIGC Illustration – Character Design White Paper (Vol. 1)' released today!** [4]: The first AIGC character visual dictionary—systematically cataloging English/Chinese terminology and illustrated examples for hairstyles, hair accessories, and apparel. - **Open-Sourced Today (May 25, 2026): Microsoft Lens officially open-sourced—a 3.8B-parameter foundational text-to-image (T2I) model** [5]: A T2I open-source foundation model featuring dense caption pretraining and mixed-resolution learning capabilities. - **Young users aren't buying heavily retouched photos—scenic-spot portrait studios are panicking** [8]: The proliferation of AI photo-editing tools is eroding information asymmetry, forcing travel photography studios to overhaul their pricing and service models. ## 🔗 Sources [0] DeepSeek adopts the 'Mixue Ice Cream & Tea' strategy to build China's answer to Claude Code — https://www.bestblogs.dev/article/5e68673c?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [2] The bottleneck preventing 60-billion-parameter LLMs from running on smartphones has finally been cracked—by a Chinese AI company — https://www.bestblogs.dev/article/1ac2cf11?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [4] Breaking! 'AIGC Illustration – Character Design White Paper (Vol. 1)' released today! — https://www.bestblogs.dev/article/6386a6d8?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [5] Open-Sourced Today (May 25, 2026): Microsoft Lens officially open-sourced—a 3.8B-parameter foundational text-to-image (T2I) model, featuring dense caption pretraining and mixed-resolution learning — https://www.bestblogs.dev/article/7b

DeepSeek launches a low-price offensive with its permanently discounted V4-Pro API, targeting the Claude Code–level programming agent market; Baishan Intelligence breaks the edge-side bottleneck by achieving 1.58-bit ternary quantization training of a 60-billion-parameter model on Huawei's Ascend platform—reducing GPU memory usage by 6× while retaining 97% of model capability [2]; meanwhile, Gemini quietly overhauls its billing logic, drastically shrinking actual usage quotas for paying users and exposing a trust gap in large-model commercialization [10].

🚀 Major Updates

DeepSeek adopts the 'Mixue Ice Cream & Tea' strategy to build China's answer to Claude Code [0]: The V4-Pro model API is permanently discounted, while DeepSeek simultaneously advances R&D of an AI programming agent tool comparable to Claude Code.
The bottleneck preventing 60-billion-parameter LLMs from running on smartphones has finally been cracked—by a Chinese AI company [2]: Baishan Intelligence, in collaboration with Tsinghua University and OpenBMB, releases BitCPM-CANN—a ternary-quantized model trained end-to-end on Huawei's Ascend platform.
Gemini has 'defected,' turning paying users into the punchline [10]: Google shifts Gemini's billing model from per-call pricing to compute-consumption-based pricing—and introduces dual usage caps—sparking widespread user backlash.
RenderFlow: Baidu's agentic code delivery in vertical search services [9]: Baidu deploys an LLM-powered agentic system, compressing code delivery cycles for individual use cases from days down to minutes.
Ant Group leads investment; Guanglun Intelligence secures $2B valuation, leading global embodied AI data infrastructure [6]: The value center of the embodied intelligence industry is rapidly shifting toward data and evaluation infrastructure.
Breaking! 'AIGC Illustration – Character Design White Paper (Vol. 1)' released today! [4]: The first AIGC character visual dictionary—systematically cataloging English/Chinese terminology and illustrated examples for hairstyles, hair accessories, and apparel.
Open-Sourced Today (May 25, 2026): Microsoft Lens officially open-sourced—a 3.8B-parameter foundational text-to-image (T2I) model [5]: A T2I open-source foundation model featuring dense caption pretraining and mixed-resolution learning capabilities.
Young users aren't buying heavily retouched photos—scenic-spot portrait studios are panicking [8]: The proliferation of AI photo-editing tools is eroding information asymmetry, forcing travel photography studios to overhaul their pricing and service models.

🔗 Sources

[0] DeepSeek adopts the 'Mixue Ice Cream & Tea' strategy to build China's answer to Claude Code — https://www.bestblogs.dev/article/5e68673c?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[2] The bottleneck preventing 60-billion-parameter LLMs from running on smartphones has finally been cracked—by a Chinese AI company — https://www.bestblogs.dev/article/1ac2cf11?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] Breaking! 'AIGC Illustration – Character Design White Paper (Vol. 1)' released today! — https://www.bestblogs.dev/article/6386a6d8?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[5] Open-Sourced Today (May 25, 2026): Microsoft Lens officially open-sourced—a 3.8B-parameter foundational text-to-image (T2I) model, featuring dense caption pretraining and mixed-resolution learning — https://www.bestblogs.dev/article/7b

← Back to Updates