日本語
2026-07-26 Morning edition
AI News Daily

AI News Daily
2026-07-26

Morning edition

Frontier models keep gaining capability without gaining price, while the largest commitments move to memory supply, data centre capacity and the rules governing AI output.

Today

Highlights

Anthropic unveils Claude Opus 5
Model release · Published: 2026-07-24
SK hynix and NVIDIA expand a multi-year AI memory and data centre partnership
Corporate · Published: 2026-07-25
Google DeepMind announces Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber
Model release · Published: 2026-07-21
FTC publishes a draft policy statement on AI accuracy and opens it for comment
Regulation · Published: 2026-07-01
OpenAI introduces Presence, an enterprise AI agent deployment platform
Corporate · Published: 2026-07-22
OpenAI makes the GPT-5.6 family (Sol, Terra, Luna) generally available
Model release · Published: 2026-07-09
OpenAI announces GPT-Live, a full-duplex voice model
Model release · Published: 2026-07-08
A retrospective edition: two items from the past 48 hours, five from the first half of this term
Editorial note · 2026-07-26
Story 01 · Model release

Anthropic unveils Claude Opus 5

Published: 2026-07-24 · Source tier: Tier 2
Context window
1M tokens
at unchanged per-token pricing
Why it matters

Major labs keep raising performance while holding prices flat, so model selection and cost estimates need re-reviewing on a schedule.

Sources: https://www.bloomberg.com/news/articles/2026-07-24/anthropic-unveils-more-cost-efficient-model-for-everyday-tasks · https://techcrunch.com/2026/07/24/anthropic-launches-opus-5/
Story 02 · Corporate developments

SK hynix and NVIDIA expand a multi-year AI memory and data centre partnership

Published: 2026-07-25 · Source tier: Tier 1
Deal scale
$500B+
AI infrastructure partnership
Why it matters

A deal at this scale resets the baseline assumptions behind corporate AI infrastructure investment and supply chain strategy.

Sources: https://news.skhynix.com/en/multi-year-tech-partnership-with-nvidia/ · https://www.cnbc.com/2026/07/25/nvidia-locks-down-memory-from-sk-hynix-as-part-of-500-billion-ai-deal.html
Story 03 · Model release

Google DeepMind announces Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber

Published: 2026-07-21 · Source tier: Tier 1 · Retrospective
Output tokens
17%
reduction in usage, same context window
Why it matters

Better low-cost, high-speed models lower the price of adopting AI for high-volume processing and automation in day-to-day operations.

Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/
Story 04 · Regulation and policy

The FTC publishes a draft policy statement on AI accuracy and opens it for comment

Published: 2026-07-01 · Source tier: Tier 1 · Retrospective
Legal basis
Section 5
of the FTC Act, on deceptive acts
Why it matters

US regulators are moving toward stronger oversight of the neutrality and accuracy of AI output, which bears directly on corporate AI governance and compliance policy.

Source: https://www.ftc.gov/news-events/news/press-releases/2026/07/ftc-seeks-public-comment-policy-statement-addressing-ai-accuracy
Story 05 · Corporate developments

OpenAI introduces Presence, an enterprise AI agent deployment platform

Published: 2026-07-22 · Source tier: Tier 1 · Retrospective
Built in
Internal policies
Permission controls
Escalation rules
Pre-deployment simulation
Why it matters

A product that goes beyond model supply into agent operations management gives enterprises a reference for deploying AI agents safely in production.

Source: https://openai.com/index/introducing-openai-presence/
Stories 06 and 07 · Model release

OpenAI ships a flagship family and a full-duplex voice model

GPT-5.6 family (Sol, Terra, Luna) becomes generally available

Published: 2026-07-09 · Tier 1
  • General availability opened on 9 July.
  • Top model Sol reaches state-of-the-art performance in coding, knowledge work, cybersecurity and science.
  • Sol became the default model in ChatGPT.

GPT-Live, a full-duplex voice model

Published: 2026-07-08 · Tier 1
  • Announced on 8 July as a new-generation voice model.
  • A full-duplex architecture lets it listen and speak at the same time as a human.
  • The stated aim is a natural conversational experience.
Why it matters

Flagship competition keeps resetting the performance-versus-cost benchmark, and more natural voice interfaces widen AI's practical range into customer support and voice agents.

Sources: https://openai.com/index/gpt-5-6/ · https://openai.com/index/introducing-gpt-live/
Overview

Three trends behind the day

Wrap-up

Three things to remember

ONE

A flagship refresh no longer means a price rise. Claude Opus 5 held its per-token pricing and Gemini 3.6 Flash cut output token usage by 17 percent.

TWO

The biggest commitment of the period was infrastructure: more than 500 billion dollars in scale, HBM4-class memory, and a 2-gigawatt data centre.

THREE

Regulators are probing how AI output is steered while vendors start selling the controls, so governance and platform choices belong in one conversation.

What to watch next

The FTC comment period closes on 31 July, and the SK Telecom AI data centre is scheduled to begin operating in 2027.