日本語
2026-07-13 Evening edition
Evening edition

AI News Daily
2026-07-13

A quiet evening for original news, but a loud week for the two forces shaping it: platform distribution, and the widening US-China split in AI compute and services.

Five stories, 2026-07-13
Today's highlights

Five stories on the evening desk

Story 01 · Regulation and policy

OpenAI and Google supplied AI services to Pentagon-listed Chinese groups

Published: 2026-07-10
  • Services reached affiliates of Alibaba, Baidu and Tencent through the two companies' Singapore-based entities.
  • All three groups sit on the US Department of Defense 1260H list of firms suspected of Chinese military ties.
  • Not illegal under current rules, but Washington's push for tighter AI-software export controls has grown louder.
Why it matters

The contradiction between global AI expansion and geopolitical risk, made concrete. Anthropic, by contrast, bars frontier models to Chinese-affiliated firms outright.

Source: OpenAI and Google Provided AI Services to Pentagon-Listed Chinese Groups via Singapore Units — Yahoo Finance https://finance.yahoo.com/technology/ai/articles/openai-google-provided-ai-services-101414934.html
Story 02 · Corporate developments

GPT-5.6 becomes the default model in Microsoft 365 Copilot

Published: 2026-07-09 to 2026-07-10
  • OpenAI will make GPT-5.6 the default across Word, Excel, PowerPoint, Chat and Cowork.
  • Follows this morning's report on general availability of the GPT-5.6 family.
  • Some outlets tied the timing to circulating reports of strain between OpenAI and Microsoft.
Why it matters

A platform's model choice lands directly in the daily work of enterprise users, and a shift in the OpenAI-Microsoft balance affects lock-in strategy for Copilot-dependent firms.

Source: OpenAI says GPT 5.6 is the 'preferred model' for Microsoft Copilot 365 amid breakup chatter — TechCrunch https://techcrunch.com/2026/07/09/openai-says-gpt-5-6-is-the-preferred-model-for-microsoft-copilot-amid-breakup-chatter/
Story 03 · Model releases

Meta ships Muse Image and the agent model Muse Spark 1.1

Published: 2026-07-07 to 2026-07-08
  • Muse Image handles complex prompts, photo compositing and QR code generation; Muse Video is in preview.
  • Muse Spark 1.1 is an agent-focused model with a one-million-token context window.
  • Zuckerberg claims parity with GPT-5.5 and Opus 4.8 on agent evaluations including MCP Atlas.
1M
token context window on Muse Spark 1.1
Why it matters

Meta now contests image, video and long-context agents at once, widening the supplier shortlist for enterprise multimodal AI.

Source: OpenAI, Meta and xAI unleash flurry of new AI models this week — Euronews https://www.euronews.com/next/2026/07/08/openai-meta-and-spacexai-push-new-ai-models-in-a-week-of-major-releases
Story 04 · Model releases

xAI launches Grok 4.5 for coding and agent workloads

Published: 2026-07-08
  • Now available through the xAI API. Musk positions it as Opus-class but faster and cheaper.
  • Pricing is $2 per million input tokens and $6 per million output tokens.
  • The underlying V9 architecture finished training on 2026-05-26; first release since the Cursor acquisition.
$2
per million input tokens (output: $6)
Why it matters

Price competition in coding and agent use intensifies, giving engineering teams reason to revisit how they weigh speed, cost and agent performance.

Source: Scoop: SpaceXAI launches new model, Grok 4.5 — Axios https://www.axios.com/2026/07/08/spacexai-grok-new-model
Story 05 · Research and infrastructure

China's fully domestic 100,000-card cluster Dengfeng goes live

Published: reported around 2026-07-12 to 2026-07-13
  • Dengfeng is reported as China's first 100,000-card AI compute cluster built entirely from domestic chips.
  • It has been completed and connected to the national supercomputing network.
  • The qualifier that carries the weight is "exclusively domestic," not the scale alone.
100,000
cards in the Dengfeng cluster
Why it matters

Evidence that China is sourcing large-scale AI compute at home despite US export controls, extending the split into the infrastructure layer.

Source: Global AI news for 2026-07-13 — eguweb https://eguweb.jp/ai/81186/
Trend overview

Aftershocks, not announcements

A thin evening for primary news

Little that was fully independent of the morning edition. Most coverage was follow-up to announcements made between 2026-07-07 and 2026-07-10.

From model launches to execution substrate

Attention is moving from standalone model releases toward implementation and platform integration, where agents carry work through to completion. The Microsoft 365 Copilot tie-in is the clearest case.

Fragmentation on both sides of the stack

The export-control dispute over AI services and China's domestic compute build-out are advancing in parallel, making the geopolitical split visible in infrastructure and in service provision alike.

Wrap-up

Three things to remember

What to watch next

Whether Washington converts the 1260H disclosure into actual export controls on AI software, and whether Meta's and xAI's performance claims hold up under independent testing.