CNBC dates the releases to a "call for slowdown"; Reuters calls the models "more accessible"

What happened. On 22 September 2026 OpenAI introduced two models, GPT-6 Sol and GPT-6 Luna, and Anthropic released Claude Opus 5.5. CNBC carries both companies in one report. The New York Times carries Anthropic alone. Reuters carries OpenAI alone.

The word is cheaper. The number is missing. All three describe the new models as costing less. None gives a price, a percentage, or the model being undercut.

Outlet	Whose release	What it calls the models	What it sets them against
CNBC	Both	"cheaper models"	"first release since call for slowdown"
The New York Times	Anthropic	"a cheaper, faster model"	"Amid Safety Debate"
Reuters	OpenAI	"mais acessíveis" — more accessible	nothing

Who speaks. Nobody outside the two companies speaks in any of the three reports. The New York Times attributes the safety claim to Anthropic and marks the testing as the company's own: a model "that it said was the strongest performer on its most rigorous internal safety tests to date."

Not settled. Who asked the industry to slow down, and when — no report we hold names them or quotes them. What either company now charges, and against what. Whether the releases answer that call, ignore it or predate it. And whether anyone outside the two firms has tested either model: no report we hold carries an outside test or a score.

This is a story about the two largest American companies selling access to artificial-intelligence models, and what it means that both were reported cutting prices on one day.

How it got here. OpenAI was founded in 2015 and released ChatGPT in November 2022, which turned language models into a consumer product and a market. Anthropic was founded in 2021 by former OpenAI staff and sells Claude. Both sell the same thing to the same buyers — developers and companies who pay by the volume of text a model reads and writes, metered per unit. That is why "cheaper" is a checkable claim in this business and an empty one without a number: the price is published as a rate, and none of the three reports gives one.

Running alongside the releases since the 2022–23 boom is an argument about whether the labs are building faster than they can test what they build. In March 2023 an open letter organised by the Future of Life Institute urged labs to pause training of the most advanced systems, and the argument has recurred at every large release since. CNBC's headline sets this week's models against a "call for slowdown." Whether that is the call it means, we cannot say — no report we hold names one.

Who is in it. OpenAI and Anthropic, each selling to the other's customers, so a price move by one is a competitive fact for the other. And the people asking for a slowdown, who are named by nobody in the reports we hold.

What is at stake. If the models really are cheaper, the cost of building AI into other companies' products falls and use spreads further. If release dates are set by what a competitor did that morning, the safety argument is that testing moves to whatever the fastest firm decides is enough.

Where it stands. Three reports, published within half an hour of each other, none longer than a sentence, none carrying a photograph.

What is being argued about

Whether the new models are cheaper, and by how much — no report prints a price
Who asked the industry to slow down, and whether these releases answer that
A company's claim about its own internal safety testing
Two firms, one release day, in a market with few sellers

One day, three reports, three different storiesSide by side

Outlet	What it said	Hedged?
"Anthropic and OpenAI roll out cheaper models in first release since call for slowdown" — both firms, parallel verbs, equal length	No. "Cheaper" is stated flat, and so is the slowdown call. Neither is sourced.	CNBC
"a cheaper, faster model that it said was the strongest performer on its most rigorous internal safety tests to date"	Yes, on safety. "It said" and "internal" both appear. "Cheaper" is not hedged.	The New York Times
"OpenAI amplia linha GPT-6 com modelos Sol e Luna mais acessíveis" — OpenAI extends the GPT-6 line with more accessible Sol and Luna models	No. "More accessible" stands in for a price, in the company's own register.	Reuters

Quotes are from the reports' own headlines and ledes. The Reuters headline is in Portuguese; the translation is ours.

Sources

CNBC

The New York Times

Reuters

The bias in this coverageBias

Cheaper is the claim; no report prints a price

All three reports say the new models cost less — cheaper, more affordable — and none gives a price, a percentage, or what the earlier models cost. One report attributes the safety claim to the company that makes it and says the tests were internal. The reports carry no figure a reader could check.

3 reports on the releases 0 reports giving a price or a percentage 1 reports attributing the safety claim to the company

The slowdown is named; the people asking for it are not

Two reports place the releases against a call to slow down or a safety debate; neither names who called for it or quotes them. The third does not mention it at all and reports the release as a product line extension. Nobody outside the two companies speaks in any of the three reports.

2 reports mentioning the slowdown call or safety debate 0 reports naming who asked for a slowdown 0 reports quoting anyone outside the two companies

Scored without the mastheads. Where the two passes differ, both are kept.

Sources

CNBC

The New York Times

Reuters

Claims and verdicts

unverified The new models from both companies cost less than the versions they follow.

CNBC calls them cheaper, the New York Times calls Opus 5.5 cheaper and faster, and Reuters calls Sol and Luna more accessible — but none of the three reports we hold gives a price, a unit or the model being undercut. Three outlets repeating an adjective is not a figure.

Checked against 3 sources:

CNBC

The New York Times

Reuters

unverified Claude Opus 5.5 is the strongest performer on Anthropic's most rigorous internal safety tests to date.

Only the New York Times carries this, and it attributes it to Anthropic, which is a company statement about its own tests of its own product. No report we hold contains an outside test, a score or the tests' content.

Checked against 1 source:

The New York Times

unverified These are the first model releases since a call for the industry to slow down.

CNBC alone frames the releases this way, and neither the New York Times nor Reuters mentions any such call; the New York Times refers only to a safety debate. No report we hold says who called for a slowdown or when.

Checked against 3 sources:

CNBC

The New York Times

Reuters

Who wrote it

E Emmy Martin The New York Times

What the bias passes found in their reporting

Attributed, not asserted the safety claim is attributed to the company and marked as its own internal testing×1

Hedged headline a debate is named in the headline and no party to it is quoted in the report×1

What is attributed the safety performance claim is attributed to the company and marked as its own internal tests×1

What the headline claims one firm in the headline, the debate named but nobody in it named×1

Claims we checked that rest on their reporting

3 unverified1 true

From 15 reports we hold under this byline, at The New York Times, The Sydney Morning Herald. Two independent readings, each made without the masthead, a quote attached to every tag. A record, never a rating. Their full record →

The bylines on the reports this story is built from. Hover one for what our passes have measured on their reporting. · 2 reports we hold carry no byline — we do not invent one.

Coverage, outlet by outlet

Reuters — independent

OpenAI amplia linha GPT-6 com modelos Sol e Luna mais acessíveis

The only report naming both new OpenAI models as an extension of an existing line, and the only one with no safety or slowdown framing at all.

What is left out: a product line extension only: cheaper is stated, the slowdown call does not appear — “OpenAI amplia linha GPT-6 com modelos Sol e Luna mais acessíveis”

Which measure is chosen: "more accessible" stands in for a price, with no figure and no comparison — “OpenAI amplia linha GPT-6 com modelos Sol e Luna mais acessíveis”

Measure named: "more accessible" stated flat, in the companies' own register, with no price attached — “OpenAI amplia linha GPT-6 com modelos Sol e Luna mais acessíveis”

Read it at Reuters → https://www.reuters.com/pt/tecnologia/42HKKG5TEBJN7ALWGBZWHIW3CI-2026-0…

CNBC — independent

Anthropic and OpenAI roll out cheaper models in first release since call for slowdown

The only report carrying both companies, and the only one dating the releases against a call for the industry to slow down.

What the headline claims: sequences the release against a call to slow down, so cheaper models read as an answer to it — “Anthropic and OpenAI roll out cheaper models in first release since call for slowdown”

Who is named as acting: both firms named as acting, in parallel verbs and equal length — “OpenAI introduced GPT-6 Sol and GPT-6 Luna. Anthropic unveiled Claude Opus 5.5.”

Framing by juxtaposition: the price cut is placed against a call for slowdown; who called is not named anywhere — “Anthropic and OpenAI roll out cheaper models in first release since call for slowdown”

Whose voice: the body is the product names and nothing else — two firms, four products, no third party — “OpenAI introduced GPT-6 Sol and GPT-6 Luna. Anthropic unveiled Claude Opus 5.5.”

Read it at CNBC → https://www.cnbc.com/2026/09/22/anthropic-openai-cheaper-ai-models.html

The New York Times — independent

Anthropic Releases a New A.I. Model, Opus 5.5, Amid Safety Debate

The only report carrying a safety claim — and it attributes it to Anthropic and says the tests were the company's own. Also the only one calling the model faster as well as cheaper.

The company released a cheaper, faster model that it said was the strongest performer on its most rigorous internal safety tests to date.

What is attributed: the safety performance claim is attributed to the company and marked as its own internal tests — “The company released a cheaper, faster model that it said was the strongest performer on its most rigorous internal safety tests to date.”

What the headline claims: one firm in the headline, the debate named but nobody in it named — “Anthropic Releases a New A.I. Model, Opus 5.5, Amid Safety Debate”

Attributed, not asserted: the safety claim is attributed to the company and marked as its own internal testing — “a cheaper, faster model that it said was the strongest performer on its most rigorous internal safety tests to date”

Hedged headline: a debate is named in the headline and no party to it is quoted in the report — “Anthropic Releases a New A.I. Model, Opus 5.5, Amid Safety Debate”

Read it at The New York Times → https://www.nytimes.com/2026/09/22/technology/anthropic-ai-model-safety…

Read another desk

Sections

Registers

Desks

CNBC dates the releases to a "call for slowdown"; Reuters calls the models "more accessible"

What is being argued about

Cheaper is the claim; no report prints a price

The slowdown is named; the people asking for it are not

Claims and verdicts

Who wrote it

Coverage, outlet by outlet