General AI News (Sep 23): GPT-6 Sol, Opus 5.5 & Trump's Rebrand
AI news for Sep 23, 2026: OpenAI cuts GPT-6 prices in half, Anthropic ships Opus 5.5, Trump rebrands AI as super intelligence, and AMD hits $1T.
Today's AI news is a price war wearing a rebrand. In the space of a few hours, OpenAI and Anthropic both shipped cheaper frontier-tier models, AMD crossed a trillion dollars in market value on the back of AI compute demand, and the White House announced it will stop calling any of it artificial intelligence. Yesterday's General AI News digest flagged an agentic security warning; today the bill came due in more ways than one.
The AI News Price War: GPT-6 Sol Meets Opus 5.5
OpenAI cuts GPT-6 token prices in half
OpenAI expanded its GPT-6 family with two cheaper tiers, GPT-6 Sol and GPT-6 Luna, and cut API pricing by 50% against the promotional rates of the previous generation. Sol drops from $4 to $2 per million input tokens and from $20 to $10 per million output tokens. Luna drops from $0.20 to $0.10 input and from $1.20 to $0.50 output. Source: OpenAI
GPT-6 Astra, released earlier in September, remains OpenAI's flagship and, per the company, its best model across the board. Sol and Luna inherit its training methods and target the everyday end of the market.
The competitive claim is unusually specific. On AutomationBench, a benchmark from Zapier that measures business workflows across connected apps, OpenAI says GPT-6 Sol running at extra-high reasoning effort beats Claude Opus 5 at maximum effort while spending 9% of Opus 5's cost per task. GPT-6 Luna at high effort improves on its predecessor by 5.4 percentage points at 58% lower cost per task. Source: VentureBeat Source: CNBC
Anthropic answers with Opus 5.5
Anthropic released Claude Opus 5.5, the successor to Opus 5 and its Opus line's cheapest release yet, at $4 per million input tokens and $20 per million output tokens. Claude Opus is Anthropic's top model tier, the one the company positions for the hardest reasoning and coding work. Source: Anthropic
CNBC noted the timing: these were the first major model releases from both labs since each published a call for a slowdown in frontier AI development. Two days later they are competing on price per token, not capability per benchmark. Source: CNBC
On the independent Artificial Analysis index (v4.3), Opus 5.5 sits at 58, comfortably ahead of the strongest open-weights challenger at 46.
Policy & Regulation
Trump says AI will be 'super intelligence' in US documents
Speaking at the United Nations, President Trump said artificial intelligence will be renamed super intelligence in all US government documents. The Hill and the Washington Post both carried the remark, and the BBC is already asking whether the rebrand will stick. Source: The Hill Source: The Washington Post
The New Republic reported that Trump credited a social media poll for the choice of words. Reuters also reported that Trump said the Justice Department could rein in AI companies if needed, which cuts against the deregulatory framing his administration has used all year. Source: Reuters
To be clear about what changed: this is a renaming of how federal documents refer to AI, not a new technical milestone or a new class of system. Nothing about model capability changed today because of the wording.
Models & Products
Xiaomi's MiMo-V2.6 goes open weights
Xiaomi released MiMo-V2.6-Pro and MiMo-V2.6 as open-weight models under the MIT license, making them free to download, modify, and deploy commercially. The Pro model is a trillion-parameter mixture-of-experts design with roughly 42 billion active parameters per token and a one-million-token context window. VentureBeat called it the top open-weights release at launch. Source: VentureBeat
Xiaomi says the model was trained with a reinforcement-learning setup spanning more than 7,000 environments for roughly $2.6 million, and demonstrated parts of the run in a public livestream. On Artificial Analysis v4.3 it scores 46, which puts it above several much larger closed models from a year ago and roughly eight points behind Opus 5.5. Source: The Next Web
Several outlets also reported that Anthropic has raised concerns that Xiaomi's model may be distilled from Claude outputs. Treat that as a developing claim: no formal finding has been published by either company, and Xiaomi has not responded publicly.
Meta's Muse lands at the top of the App Store
Bloomberg reported that Meta Muse, Meta's new AI assistant app, climbed to the top of the US App Store charts within three days of launch. Source: Bloomberg
Muse is a consumer assistant built around Meta's own models and priced at $20 a month or $100 a year. The interesting wrinkle came from the other direction: Amazon has moved to block Meta's Muse agent from shopping on its platform, a first sign that agent access, not model quality, is becoming the real battleground for consumer AI. Source: Forbes
Chips, Data Centers & Energy
AMD joins the $1 trillion club
AMD became the latest company to cross a $1 trillion market capitalization, with the stock up roughly 75% over the past month and about 150% year to date, driven by demand for its Instinct MI450 data-center accelerators. MI450 is AMD's flagship AI accelerator line, the hardware it sells into multi-gigawatt training and inference buildouts that were previously an Nvidia monopoly. Source: Reuters
The contracts behind that run are large and already on the books: multi-gigawatt compute agreements with OpenAI and Anthropic signed earlier this year. AMD's trillion is the market pricing in capacity that ships in 2027, not revenue already earned.
Security
An AI-fueled hack hit 100 companies in days
Forbes reported that a massive AI-fueled hacking campaign reached 100 companies in a matter of days, with credit card data among the haul. Source: Forbes
This is not an isolated story. Reuters reported that Anthropic disclosed a fourth AI hacking incident that its earlier review missed, TechCrunch reported that researchers used Anthropic's Claude to hack into OpenAI, and the Washington Post reported that the hackers who broke into OpenAI warn the industry has a security problem. The common thread across all four: agents doing reconnaissance and exploitation at machine speed, against companies that assumed human-paced defence was enough. Source: Reuters Source: TechCrunch
Details of the 100-company campaign are still emerging, so treat specific cost and count figures as preliminary.
Quick Comparison
| Entity | Key Metric | What Changed | Source |
|---|---|---|---|
| OpenAI GPT-6 Sol | $2 in / $10 out per million tokens | New mid tier, 50% price cut | OpenAI |
| OpenAI GPT-6 Luna | $0.10 in / $0.50 out per million tokens | New low tier, 50% price cut | OpenAI |
| Anthropic Opus 5.5 | $4 in / $20 out; 58 on AA v4.3 | New top-tier model at lower price | Anthropic |
| Xiaomi MiMo-V2.6-Pro | 1T params, 46 on AA v4.3 | Open weights under MIT license | VentureBeat |
| AMD | $1T market cap | Stock up ~75% in a month | Reuters |
| Meta Muse | No.1 US App Store, $20/month | Consumer agent launch and Amazon block | Bloomberg |
Frequently Asked Questions
What is the top AI news today?
The top AI news today is OpenAI and Anthropic shipping cheaper frontier models within hours of each other, with OpenAI cutting GPT-6 Sol and Luna API prices by 50% and Anthropic launching Opus 5.5 at $4 per million input tokens. Both releases came two days after each lab called for a slowdown in frontier AI.
Why did Trump rename AI to super intelligence?
Trump said the United States will use the term "super intelligence" instead of artificial intelligence in all federal documents, a remark he made during his United Nations address. The New Republic reported he credited a social media poll for the wording. It is a vocabulary change, not a new model or capability.
How good is Xiaomi's MiMo-V2.6-Pro open model?
MiMo-V2.6-Pro scores 46 on the Artificial Analysis index while shipping under the MIT license, which puts it close to the strongest closed models of a year ago. Xiaomi reports roughly one trillion parameters, 42 billion active, a one-million-token context window, and about $2.6 million in training cost.
Bottom Line
Two labs cut prices on the same day, a trillion dollars of chip demand got repriced, and the US government decided the words matter enough to change. The story underneath the price war is the one to watch: if frontier capability is now a commodity sold at $2 per million tokens, the durable advantages shift to distribution, agent access, and data. More AI news and the rest of this week's coverage are on the blog.
Related reading
Background worth having before you act on the above:
- General AI News (Sep 30): GPT-6.1 Sol, Dots & Trump
- Anthropic risk report, UK frontier-AI dependency check, GPT-5.6 Sol at half price, China cost war: General AI News for August 18, 2026
- Viral AI News (Sep 25): Opus 5.5 Video, Orbit TPUs & $500 Leak
- Viral AI News (Sep 15): RTX 5090 Gone, Trump & CrofAI
Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →
