日本語
2026-07-11 Morning edition
AI News Daily
Morning edition
AI News Daily
2026-07-11
Capability, capital and regulation all moved in the same week: three new frontier models, a 14-gigawatt compute plan, and governance debates opening at both the UN and Japan's governing party.
Seven stories · Model releases, corporate moves, regulation and policy, Japan
Today's highlights
| Category | Headline | Published |
|---|
| Model release | OpenAI makes GPT-5.6 (Sol, Terra, Luna) generally available and the new ChatGPT default | 2026-07-09 |
| Model release | xAI ships Grok 4.5 at a low price point and ranks fourth on the intelligence index | 2026-07-08 to 07-10 |
| Corporate | Meta starts custom AI chip production and launches Muse Spark 1.1 | 2026-07-09 to 07-10 |
| Corporate | AI unicorn Mercor acquires the startup Deeptune | 2026-07-09 |
| Corporate | Anthropic reported at roughly $47.7 billion annualized revenue | 2026-07-08 to 07-09 |
| Regulation | UN panel warns catastrophic harm cannot be ruled out, urges stronger governance | 2026-07-06 to 07-07 |
| Japan | LDP proposes expanding the Cabinet Office AI policy unit to more than 80 staff | 2026-07-09 |
OpenAI releases the GPT-5.6 family
Published: 2026-07-09 · Model release
- Sol, Terra and Luna went to general availability on 9 July and became the new default model in ChatGPT.
- Flagship Sol scored 88.8% on Terminal-Bench 2.1, and 91.9% in ultra mode.
- That edges past the previous generation, GPT-5.5, at 88.0% on the same benchmark.
88.8%
Sol on Terminal-Bench 2.1
(91.9% in ultra mode)
Why it matters
Because it is the ChatGPT default, the upgrade reaches every user at once. Anyone using ChatGPT for coding or agent work should feel the accuracy gain without changing anything.
xAI ships Grok 4.5 at a low price point
Published: 2026-07-08 to 2026-07-10 · Model release
- Grok 4.5 became generally available on 8 July as a coding-focused model.
- Pricing is $2 per 1M input tokens and $6 per 1M output tokens.
- It scored 54 on the Artificial Analysis Intelligence Index, fourth behind Claude Fable 5, GPT-5.5 and Claude Opus 4.8.
54
Artificial Analysis
Intelligence Index, 4th place
Why it matters
Near-frontier quality at a fraction of the token cost widens the options for enterprises running high-volume API workloads, where the monthly bill outweighs a few index points.
Meta builds its own silicon, and a coding-agent model
Published: 2026-07-09 to 2026-07-10 · Corporate
- Custom AI chip manufacturing begins in September.
- Compute capacity is planned to reach 14 gigawatts by 2027.
- Muse Spark 1.1, a model for coding agents, launched on 9 July.
14 GW
Planned compute capacity
by 2027
Why it matters
Expanding in-house chip investment shows Meta moving away from external GPU dependence toward a vertically integrated AI stack, which reshapes the industry-wide contest for compute.
Consolidation and revenue on the corporate side
Published: Mercor 2026-07-09 · Anthropic 2026-07-08 to 2026-07-09
- Mercor, an AI talent-matching company valued at around $10 billion, acquired the startup Deeptune, in which founder Brendan Foody had invested. Backers include a16z, OpenAI and Anthropic.
- Anthropic's annualized revenue was reported at roughly $47.7 billion, alongside the rollout of Claude Cowork for mobile.
$47.7B
Anthropic annualized
revenue, as reported
Why it matters
Rapid revenue growth confirms strong enterprise demand for AI agents, while the acquisition shows an ecosystem reorganising itself as funding stays hot.
UN panel urges stronger international AI governance
Published: 2026-07-06 to 2026-07-07 (UN reporting: 07-10) · Regulation and policy
- The UN Global Dialogue on AI Governance was held in Geneva on 6 and 7 July.
- The UN independent scientific panel, led by figures including Yoshua Bengio, warned that AI capability is outpacing policy and scientific understanding.
- The panel said the possibility of catastrophic harm cannot be ruled out, and called on states to strengthen the international regulatory framework.
Why it matters
A serious AI governance debate at UN level, with the major states present, can shape the direction of future international frameworks such as the EU AI Act.
Japan: LDP calls for a bigger AI policy office
Published: 2026-07-09 · Japan
- On 9 July the AI and web3 subcommittee of the LDP's Headquarters for the Promotion of a Digital Society, chaired by Masaaki Taira, presented a draft emergency proposal.
- It asks for the Cabinet Office AI Policy Promotion Office to grow from just under 30 staff to more than 80.
- It also covers high-salary recruitment of private-sector AI specialists and a new AX promotion allocation in the fiscal 2027 budget.
80
Staff sought for the Cabinet
Office AI policy unit
Why it matters
Strengthening the machinery that delivers Japan's AI policy directly affects how fast domestic regulation and industrial support measures can move.
Trend overview
Models on top of each other
OpenAI, xAI and Meta all launched in nearly the same window with GPT-5.6, Grok 4.5 and Muse Spark 1.1. Competition on both price and performance has stepped up another level.
Record funding, active M&A
Fundraising and acquisitions stay at record levels. The ecosystem is consolidating, as with Mercor and Deeptune, while major players such as Anthropic show sharply expanding revenue.
Governance in parallel
The UN Global Dialogue and Japan's proposal to reinforce its government apparatus show governance activity rising at both international and domestic levels, alongside the model race rather than after it.
Wrap-up
Three things to remember
- The leaderboard is crowded: Sol at 88.8% versus GPT-5.5 at 88.0% means price and availability now decide, not raw score.
- Money is following usage: Anthropic at about $47.7 billion annualized, Meta reinvesting into 14 gigawatts of its own compute.
- Rules are no longer trailing: the UN in Geneva and the LDP in Tokyo both moved on AI governance in the same week.
What to watch next
Whether cheap near-frontier models such as Grok 4.5 pull enterprise workloads away from the premium tier, and whether Japan's fiscal 2027 budget process turns the LDP proposal into an actual expansion of the Cabinet Office AI unit.