Recent AI news · Original report: 2026-10-06 (KST)

Gemini 4 Argon release date and pricing, ChatGPT image ads, text watermark (Oct 6, 2026)

This article was compiled with AI. This English translation preserves the original report's collection date, findings, and limitations. Please consult the linked sources for current information. Originally collected: 2026-10-06 10:10 (KST, UTC+09:00).

Translated into English: 2026-10-06 (KST). The translation date does not indicate a new source check.

At a glance

This week's three announcements are one new model and two sets of rules. Google's new frontier model arrived as a name and a price first, OpenAI said it will test visual ads in ChatGPT, and ChatGPT text in the European Union will carry a watermark. None of the three applies to users in Korea right away.

This week's announcements

We opened the official posts ourselves, starting at 10:10 a.m. Korea time on October 6, 2026. Dates are the ones printed on the original posts.

Gemini 4 Argon: announced, but not yet available

Google announced its new frontier model, Gemini 4 Argon, on September 30. For now it is rolling out only to trusted cyber defenders in the Fairwind Program.

• Public release: There is no date. Google wrote that it will gather feedback from early testers, refine its guardrails, and then open the model starting with paid API customers and Google AI Ultra subscribers.
• Pricing: The introductory price is $2 per million input tokens and $10 per million output tokens. A footnote says the price becomes $4 for input and $20 for output once the introductory period ends. The post does not say when that is.
• What changed: The output limit for a single response grows from 64K tokens to 1 million tokens.
• Scores Google cited: 77.9% on DeepSWE v1.1, which measures long software engineering tasks, and 91.7% on LVBench, which measures long video understanding. Both figures come from Google's post.
• Why the wait: Google said it is strengthening four areas before a broad release: defending against misuse, defending against prompt injection, monitoring whether the model goes beyond what the user intended, and hardening its test environments. It is also taking part in the U.S. government's voluntary process for pre-release model access.

ChatGPT visual ads: a US test later this month

On October 5, OpenAI said it is bringing a new image-based ad format to ChatGPT.

• Where it appears: The first test shows the ads while ChatGPT is generating an image.
• Principles OpenAI stated: Ads will be clearly labeled and kept separate from the image being created. OpenAI said advertising does not influence ChatGPT's answers.
• When and where: Testing begins later this month in the US with an initial group of advertisers.
• When ads stay away: OpenAI wrote that it evaluates conversations so ads do not appear when a user is emotionally vulnerable or the topic is sensitive.
• Scale: OpenAI said 1.2 billion people use ChatGPT each week.

A watermark for AI-written text, starting in the EU

On the same day, OpenAI said it will phase in text watermarking in response to the EU AI Act. The technology, called textGrain, leaves an invisible statistical signal in the model's word choices.

• Where: Over the coming weeks it will be added to ChatGPT and Codex text in the European Union, across all plans. It will not be a global default. API customers have been able to opt in since October 5, and it is off by default.
• The detector: Access is limited to approved researchers and expert organizations. OpenAI said it is not making the detector public because of the risk of missed watermarks and false positives.
• Accuracy in OpenAI's own tests: At a 1% false positive rate, the detector found the watermark in about 95% of 400-token psychology passages and about 80% of 200-token passages. In 400-token passages, replacing 10% of the words with synonyms cut detection from about 92% to 66%, and replacing 25% cut it to 17%.
• What the watermark does not tell you: It does not show how much a person edited the text, which user generated it, or whether the content is accurate. OpenAI also wrote that the absence of a watermark is not proof that a human wrote the text.

What it means and for whom (our estimate)

• A tool that settles whether a child's homework or a cover letter was "written by AI or not" still looks out of reach. The company that built the watermark says it often misses short or edited text, and ChatGPT text written in Korea does not get the watermark by default.
• The ad test is limited to the US, so screens in Korea are unlikely to change right away.

What we could not confirm

• Not tested by us: Argon is not open to the public, and we have not seen the ads or the watermark ourselves. The scores and detection rates all come from the companies that announced them.
• Korea: The post does not say when ads will arrive in Korea or which plans will see them.
• Legal provisions: OpenAI's post does not give the article number or effective date in the EU AI Act, and we did not open the text of the law.
• Argon price list: The Gemini API pricing page had no Argon entry when we checked. The prices are those in the announcement post.

Sources

• Google Blog, Gemini 4 Argon: our next era of frontier intelligence (2026-09-30) https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
• Google Blog, The latest AI news we announced in September 2026 (2026-10-02) https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-september-2026/
• Google, Gemini API pricing documentation https://ai.google.dev/gemini-api/docs/pricing
• OpenAI, Building advertising for the way people use AI (2026-10-05) https://openai.com/index/new-chatgpt-ads-format-and-measurement
• OpenAI, Our approach to EU text provenance rules (2026-10-05) https://openai.com/index/eu-text-provenance