Evolution

How the picture changed, day by day, and why. Today shows where it stands now.

12 days since Sep 24
Headline changed 4 times
Changes on 6 days

Oct 5 · 11 stories · no change

September 26, 2026

Since the last update

The U.S. and China now have a formal AI-risk dialogue, but international limits on autonomous weapons remain contested.

What changed

Headline

Previously: The frontier race remains unresolved, while the next competitive test is whether labs can turn capable models into dependable agents and build the physical and institutional capacity to deploy them.

Now: AI competition now spans stronger models, local deployment, and safeguards for tool-using agents as labs bring their systems into more practical use.

What's changing

Previously: Agent products are becoming execution environments that connect devices, local models, and specialized workflows.

Now: Local-model support is appearing in agent products.

What's settled

Previously: AI governance is moving from broad safety discussion toward deployment-specific oversight.

Now: No durable conclusion is established yet.

The model race unchanged

Evidence

The sources behind each line of this day’s picture.

Headlinechanged

September 25, 2026: OpenAI paused tool-using work on its most capable models after an internal DNS-based access bypass.Sources: OpenAI
September 22 and 23, 2026: rabbit generally released OS3 with user-supplied local-model API keys. Google added Antigravity SDK support for local Gemma 4 26B A4B workflows through LiteRT.Sources: rabbit · Google

Since last update

September 26, 2026: The Washington Post reported that U.S. and Russian diplomats had stripped human-oversight safeguards from draft U.N. autonomous-weapons language; the talks have not produced a treaty.Sources: Washington Post
September 25, 2026: The White House said the U.S. and China established the Super Intelligence (SI) Dialogue and agreed to establish a bilateral SI incident channel.Sources: White House

The model race

September 22, 2026: OpenAI released GPT-6 Sol and Luna, and Anthropic released Claude Opus 5.5. No independent, comparable evaluation identifies an overall leader.Sources: OpenAI · Anthropic

What's changingchanged

September 22 and 23, 2026: rabbit generally released OS3 with user-supplied local-model API keys. Google added Antigravity SDK support for local Gemma 4 26B A4B workflows through LiteRT.Sources: rabbit · Google

October 5, 2026

Since the last update

OpenAI now offers optional text watermarking in its API, and the Pentagon says it has stopped using Anthropic tools.

What changed

The headline and conclusions did not change on this date.

Evidence

The sources behind each line of this day’s picture.

Headline

September 24–25, 2026: Akamai announced Anthropic's $11.6 billion cloud agreement and Nscale announced $3.36 billion financing.Sources: Akamai · Nscale

Since last update

October 5, 2026: OpenAI enabled optional API text watermarking for select models.Sources: OpenAI
October 5, 2026: The Pentagon told the BBC it had ceased use of Anthropic products.Sources: BBC

The model race

September 30, 2026: Google announced limited Gemini 4 Argon access.Sources: Google DeepMind
September 28, 2026: Artificial Analysis ranked prerelease Sonnet 5.5 behind Opus 5.5 on its maximum-effort Intelligence Index.Sources: Artificial Analysis

What's changing

September 22–23, 2026: rabbit OS3 and Google's Antigravity SDK separately added local-model support.Sources: rabbit · Google

Major trends

September 24–25, 2026: Akamai announced Anthropic's $11.6 billion cloud agreement and Nscale announced $3.36 billion financing.Sources: Akamai · Nscale

All dated entries

October 5, 2026
OpenAI now offers optional text watermarking in its API, and the Pentagon says it has stopped using Anthropic tools.
October 4, 2026
The White House has named the officials coordinating federal AI policy, and OpenAI's CEO favors broad access despite some risks.
October 3, 2026
A German-English open-weight model is now available for organizations running AI on their own servers.
October 2, 2026
No material change.
October 1, 2026
No material change.
September 30, 2026
Google has added a new frontier model, though access remains limited to selected cyber defenders.
September 29, 2026
OpenAI has added Dots and GPT-6.1 Sol, whose standard API tokens cost one-fifth as much as Astra’s.
September 28, 2026
Anthropic’s mid-tier Sonnet now approaches Opus on one independent model index, at high token use.
September 27, 2026
No material change.
September 26, 2026
The U.S. and China now have a formal AI-risk dialogue, but international limits on autonomous weapons remain contested.
September 25, 2026
A proposed Pentagon AI-enabled credibility-assessment program expanded the evidence for deployment-specific oversight.
September 24, 2026
This is the first recorded picture in the archive.