Last checked October 27, 2026 · News roundup
Today in brief: OpenAI cut the price of its everyday model in half with GPT-6.1 Terra, Anthropic made its IPO filing public and picked the Nasdaq ticker ANT, and iOS 27.2 gave Siri a memory of past chats. On the business side, Microsoft offered voluntary buyouts to about 60,000 staff and Meta began a $95 billion stock sale to pay for AI data centers.
Monday was busy, so this edition sticks to stories dated October 26 unless noted. If you build with coding agents, read the AGENTS.md security item before you open your next project from the Downloads folder.
Part 1 · AI
GPT-6.1 Terra price: $1 per million input tokens, half of GPT-6.1 Sol
OpenAI released GPT-6.1 Terra on Monday, October 26, as the successor to GPT-5.6 Terra. In the API, the model ID is gpt-6.1-terra and it costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol.
In ChatGPT, the launch post lists Plus, Pro, Enterprise and Edu users, inside ChatGPT Work and Codex. Free and Go plans are not named. OpenAI says Terra's usage allowance covers more tasks than GPT-5.6 Terra and substantially more than GPT-6.1 Sol or GPT-6 Astra, so heavy users should hit limits later.
The pitch is computer use at a lower price: reading what is on screen, filling forms and acting across apps. OpenAI's own numbers show its error rate on trick prompts at low reasoning effort falling from 53.6% to 16.7%, and its Terminal-Bench 4.0 score at maximum effort rising to 50.3% from 21.5%. OpenAI still recommends Sol for the hardest coding work.
What to do: if you pay for GPT-5.6 Terra in the API, run your own prompts against the new model first. The bill should halve at the same token volume, but output quality on your tasks is the test that matters.
GPT-6.1 Terra API price per 1M tokens
iOS 27.2 Siri memory: what the update adds and which iPhones get it
Apple released iOS 27.2 on Monday, October 26, the first big update to iOS 27. The headline change is that Siri can now remember earlier conversations, so you can ask about something it told you last week, such as a restaurant suggestion, according to MacRumors.
Siri's AI features can also pull from photos sent in Messages, the people on a Calendar event, your voicemails, and your Safari history and Reading List. A new Excluded Apps setting lets you choose, in one place, which apps Siri, Search and Suggestions may read. Siri AI adds French, Japanese, Korean, Portuguese and Spanish.
The Siri AI features need an iPhone 15 Pro, iPhone 15 Pro Max or any iPhone 16 or later, and they are not offered in the EU or mainland China. The update itself installs on every iPhone that runs iOS 27. Photos and Image Playground also gain SynthID watermarks that help mark images made or edited with generative tools.
What to do: update in Settings, then General, then Software Update. Before you start leaning on Siri's memory, open the new Excluded Apps setting and remove anything you would not want an assistant to quote back to you.
Can an AGENTS.md file hijack Codex or Claude Code? Researchers say yes
A planted instructions file can steer a coding agent into installing malware, according to research from Forever Security reported by The Hacker News on October 26. Coding agents read AGENTS.md or CLAUDE.md files in the folder where they start, and a web page can drop a file into Downloads without a click.
In tests, a fake AGENTS.md dressed up as setup notes told OpenAI's Codex to register a server in ~/.codex/config.toml and install the npm package function-color, part of a malware chain researchers call MALFEX. Codex ran it in more than 90% of attempts. A CLAUDE.md variant tried to get Claude Code to write a new ANTHROPIC_BASE_URL, which would send API traffic and the login token to an attacker's server. Claude Code asks for approval and often refuses, but the longer, engineering-style version was approved about half the time.
Forever says neither vendor has anything to patch, because the agents behaved as designed. The habits we suggest for reviewing a third-party Claude skill apply here: read instruction files you did not write.
Five checks before your next coding-agent session
Gemini 4 Argon for Android coding: fifth on Google's own Android Bench
Google refreshed its Android Bench leaderboard on October 26, and its newest model did not top it. According to 9to5Google, Gemini 4 Argon ranks fifth with a 21.3% pass rate, behind Claude Sonnet 5.5 in first place at 40% and GPT-6.1 Sol at 29.3%. Claude Opus 5.5 sits second.
Cost cuts the other way. Google's average cost per run is $99.50 for GPT-6.1 Sol, $267.70 for Argon and $1,574.90 for Sonnet 5.5, so the top scorer is also by far the most expensive to run on this test. Completion rates for all three sit between 80.9% and 82.8%.
This is an update to our earlier guide on Gemini 4 Argon pricing and plan access. If Android apps are your main workload, the leaderboard is worth a look before you standardize on one model.
Android Bench pass rate vs. average cost (October 26 update)
Did OpenAI disprove the Exponential Time Hypothesis? The claim is now under review
OpenAI said on October 26 that an internal model produced a refutation of the Exponential Time Hypothesis (ETH), a conjecture about how hard the Boolean satisfiability problem is that hundreds of computer science papers rely on. OpenAI says the model found it on October 16 and the proof was checked in the Lean proof language the next day, with both posted on GitHub.
It is a claim, not a settled result. Russell Impagliazzo and Ramamohan Paturi, two of the researchers who proposed ETH, and MIT's Ryan Williams will lead an open review, and they said their involvement is not an endorsement, according to The Next Web. OpenAI also claimed results on the local smoothing conjecture and on Yang-Mills theory, and gave $5 million to the Simons Foundation for workshops on AI-produced results.
For ChatGPT users nothing changes today. For anyone who relies on hardness assumptions in security or algorithm work, watch the review rather than the headlines.
Part 2 · SaaS & business
Anthropic IPO: ticker ANT on Nasdaq, and what the public S-1 adds
Anthropic made its IPO prospectus public on Monday, October 26, and plans to list on the Nasdaq under the ticker ANT, according to CNBC and VentureBeat. We covered the reported draft numbers in our October 21 roundup; here is what is new.
The filing confirms first-half 2026 revenue of $16.3 billion, up from $1.16 billion a year earlier, with usage-based sales making up 80%. More than 1,900 customers paid at least $1 million over the 12 months to September 30, and about 250 paid at least $10 million. The operating loss narrowed to $693 million, and operating cash flow was a positive $3.93 billion. Most of the $80.7 billion net loss came from accounting changes on convertible notes. The S-1 also discloses an FTC civil investigative demand issued in October; that is not a finding of wrongdoing.
No share count, price range or trading date has been set. Separately, Bloomberg reported that Broadcom arranged a $63 billion financing vehicle to buy chips and lease them to Anthropic. Claude plans and prices do not change because of the filing. This is news, not investment advice.
Anthropic's public S-1 at a glance
$16.3B
Revenue, first half of 2026 (vs $1.16B a year earlier)
1,900+
Customers paying $1M+ in the 12 months to Sept 30
~250
Customers paying $10M+ in the same period
$693M
Operating loss, H1 2026 (vs $2.14B a year earlier)
$3.93B
Positive operating cash flow, H1 2026
ANT
Planned Nasdaq ticker; no price range or date yet
Microsoft buyouts: about 60,000 employees offered a voluntary exit
Microsoft confirmed to TechCrunch on October 26 that it is offering voluntary buyouts to around 60,000 employees at vice president level and below on selected teams in the US, Canada, India and the UK. Bloomberg reported that Xbox, Windows, Copilot and Finance are among the participating groups, while some data center, sales and marketing staff are excluded.
Terms are similar to Microsoft's standard severance. Chief people officer Amy Coleman wrote that the offer "gives eligible employees who may be considering a future outside the company a new choice," and Microsoft called the program new for the company. For scale, Microsoft had about 223,000 employees at the end of June.
For Microsoft 365 and Copilot customers there is no product or pricing change attached. The practical signal is cost pressure: Microsoft reports earnings this week, and it has already been moving Copilot toward usage-based billing.
Meta $95 billion stock sale: how the AI fundraising works
Meta started raising $95 billion in new equity on October 26 to fund AI infrastructure. The plan has two parts: $35 billion through underwritten public offerings and $60 billion through an at-the-market program, which lets Meta sell shares gradually at market prices. Nvidia, AMD and Broadcom are cornerstone investors, according to The Irish Times, and TheStreet reported they showed interest in buying up to $5 billion of shares together.
Meta has projected capital spending of up to $145 billion this year and has warned investors it will need to raise "hundreds of billions of dollars" over several years. TheStreet reported Meta shares were down 1.7% in early trading Monday.
For advertisers and businesses using Meta's tools, nothing changes today. It does explain why big platforms keep pushing paid AI features: the hardware bill is now large enough to need outside money even at Meta's size.
Proton AI re-interviews: staff reassessed November 4 to 11
Proton, the company behind Proton Mail, Proton VPN and Proton Drive, told staff it will "re-interview" employees across the company about how they use AI at work, according to an internal email reported by 404 Media on October 26. The email says the sessions run November 4 to 11, may inform decisions about "continued employment," and that "these re-interviews are consequential."
Interviewers will come from outside each person's team, drawn from leaders the email describes as showing strong AI ability. Proton says it has already interviewed more than 40 senior leaders in engineering and product. Not everyone will be re-interviewed, and Proton did not respond to 404 Media's request for comment.
Proton has announced no product, price or privacy change linked to this. Paying customers do not need to do anything, but it is another sign that SaaS companies now treat AI fluency as a job requirement rather than a perk.
What people are asking about today's news
Is GPT-6.1 Terra available on the free ChatGPT plan?
OpenAI's launch post lists Plus, Pro, Enterprise and Edu users, inside ChatGPT Work and Codex, and does not mention Free or Go plans. Developers can use it in the API as gpt-6.1-terra at $1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol.
When will Anthropic stock start trading?
No date has been set. Anthropic's public S-1, filed October 26, names Nasdaq and the ticker ANT but leaves the share count and price range blank. Those usually arrive in an amended filing shortly before a roadshow, and trading begins after the offering is priced. Watch the SEC's EDGAR system for the update.
Which iPhones get Siri memory in iOS 27.2?
The Siri AI features, including memory of past conversations, need an iPhone 15 Pro, iPhone 15 Pro Max or an iPhone 16 or later. iOS 27.2 itself installs on every iPhone that supports iOS 27, but Siri AI is not available in the EU or mainland China, according to MacRumors.
Is it safe to run Claude Code or Codex in my Downloads folder?
Researchers advise against it. Forever Security showed that a web page can drop an AGENTS.md or CLAUDE.md file into Downloads, and agents follow those instructions when they start there. Work from a clean project folder, read instruction files you did not write and keep approval prompts switched on.
Saazora take: The cheapest story of the day is also the most useful one. A model that halves your API bill and a ten-minute folder check against planted agent files will do more for most small teams this week than any IPO headline. For more on running agents safely, browse our guides to Claude Code and other AI agents.
Sources: OpenAI, Introducing GPT-6.1 Terra (Oct 26) · 9to5Mac, GPT-6.1 Terra (Oct 26) · MacRumors, iOS 27.2 (Oct 26) · The Hacker News, malicious AGENTS.md (Oct 26) · 9to5Google, Android Bench (Oct 26) · The Next Web, OpenAI ETH claim (Oct 26) · VentureBeat, Anthropic S-1 (Oct 26) · TheStreet, market report (Oct 26) · Bloomberg, Broadcom financing (Oct 26) · TechCrunch, Microsoft buyouts (Oct 26) · The Irish Times, Meta stock offering (Oct 26) · 404 Media, Proton re-interviews (Oct 26). All 2026.
