@ManhTran.
All posts
AI

The AI race heats up in July 2026: China's GLM-5.2 shakes things up

@M@ManhTranJuly 13, 20261 min read23 reads
The AI race heats up in July 2026: China's GLM-5.2 shakes things up

A Chinese model six times cheaper yet nearly as capable as America's best is putting Silicon Valley on edge — just as Anthropic, OpenAI, Meta and Google rush out new products.

In the second half of 2026, the artificial-intelligence race entered a fierce acceleration phase. In just the first few weeks of July, a wave of new models and features launched — but the biggest talking point came from China.

GLM-5.2: a "mini DeepSeek moment"

Beijing startup Z.ai (Zhipu) released GLM-5.2 — an open-source model whose coding and agentic capabilities are rated close to America's top models, but at roughly one-sixth the cost. The tech world is calling it a "mini DeepSeek moment."

  • Z.ai claims performance nearing Claude Opus 4.8 (Anthropic) and GPT-5.5 (OpenAI).
  • On the OpenRouter coding platform, GLM-5.2 ranked above Anthropic's models; it placed 5th on the Artificial Analysis leaderboard.
  • A 1-million-token context window (holding ~750,000 words in "working memory"), open source, with no regional restrictions.

What unsettles U.S. labs isn't the current numbers, but the speed at which the gap is closing — and a price point that breaks the market norm.

America's giants strike back

Under that pressure, the leaders kept launching:

  • Anthropic — Claude Sonnet 5 (June 30, 2026): the most "agentic" Sonnet model yet, with performance nearing Opus 4.8 but far cheaper (introductory pricing of $2 per million input tokens, $10 output). It can plan on its own, use tools (browser, terminal) and check its own results.
  • OpenAI — GPT-5.6: its public rollout was delayed after the U.S. government requested early access and additional security oversight before a wide release.

Not just chatbots: images and video

The new wave spread into multimedia content creation too:

  • Meta — Muse (July 7): an image-generation model built into Meta AI that restores old photos, changes styles (Renaissance-style portraits, claymation), builds product shots, and more.
  • Google — Video Remix (July 8): a Google Photos feature using the Gemini Omni model that turns ordinary videos into stylized clips in seconds.

The money keeps pouring in

Behind the technology race is enormous capital: Blackstone committed $30 billion for AI data centers in Japan; Qualcomm acquired AI startup Modular for $4 billion.

What does this mean?

For users and businesses, the good news is clear: fierce competition is driving AI costs down while capabilities rise fast. The rise of cheap, open models like GLM-5.2 is reshaping the rules — this is no longer the private playground of a few U.S. labs. The second half of 2026 promises more surprises, and the line between "leader" and "chaser" is thinning month by month.

References

Share this article

manhtran.tech/blog/cuoc-dua-ai-thang-7-2026-glm-5-2

Comments (0)

No comments yet. Be the first to share your thoughts!