Last checked November 12, 2026 · News roundup
Today in brief: OpenAI's model contract with Cursor reaches its proposed end date today, so developers who pick GPT models inside the SpaceX-owned editor should check what still works. ChatGPT is dropping its two older voice modes on December 11, and Anthropic handed paid Claude subscribers one free reset of every usage limit. On the business side, Railway put a price on its October breach ($740,000 in fraudulent AI charges), and Nvidia's board is reportedly preparing to replace Jensen Huang.
Deadlines from today's news, in order
Nov 12
Cursor: OpenAI's proposed last day of direct model supply
Nov 13
Anthropic IPO: livestreamed investor event with Dario Amodei and Jamie Dimon
Dec 10
Claude: free full usage reset must be used before 2:00 p.m. PT
Dec 11
ChatGPT: Advanced and Standard voice retired; users move to GPT-Live
Section 1 AI models, apps and features
Can you still use GPT models in Cursor? OpenAI's contract ends November 12
Today is the date OpenAI proposed for ending direct supply of its models to Cursor, the AI code editor that SpaceX bought. According to Cyber Security News, SpaceX closed its roughly $60 billion purchase of Cursor's parent, Anysphere, in mid-August, and OpenAI used the change-of-control clause in its contract two weeks later. OpenAI said it "cannot be confident that SpaceX will use our technology within our terms of service," as quoted by Superframeworks. TokenPost reported on November 11 that the official termination date still needs confirmation from both companies.
Who is affected: anyone who manually selects GPT models in Cursor. Cursor CEO Michael Truell has said OpenAI models make up about 5% of the editor's AI traffic, and models from Anthropic and Google remain on the menu. Reports disagree on the details: TokenPost says some features may still reach OpenAI through your own API key, the Codex extension or a gateway, while other coverage says Cursor's built-in features that relied on GPT will stop. Using your own key means paying OpenAI's per-token rates on top of your Cursor plan.
What to do today: open Cursor's model settings and confirm which model each saved agent or automation uses, search your repos for hardcoded GPT model names, and test one request with your own OpenAI key before you rely on it. If you are rethinking your stack, our guide to AI app builders and coding tools covers the main options.
ChatGPT Advanced Voice is going away on December 11: what GPT-Live changes
OpenAI will retire Advanced voice and Standard voice in ChatGPT on December 11, 2026, and move everyone to GPT-Live, the voice model it introduced in July. The OpenAI Newsroom account said voice usage has nearly doubled since then and that virtually all voice users already use GPT-Live. Notices went out on November 10, and the two older options will disappear from Settings > Voice after the cutoff.
GPT-Live can listen and talk at the same time, stays quiet while you think and hands harder questions to OpenAI's text models without ending the call. Two things stay on Advanced voice for now: live video and screen sharing, which GPT-Live does not support yet, and voice chats with custom GPTs until December 11. OpenAI gave no timeline for adding video to GPT-Live and announced no price change.
If you built a workflow around custom GPT voice or screen sharing, test GPT-Live now so you have four weeks to adjust rather than finding out on December 11.
Claude free usage reset: who gets it and when it expires
Anthropic gave every Claude Pro, Max, Ultra and Team subscriber one free full reset that covers all models and all limits. The ClaudeDevs account posted late on November 10 that it can be used any time before December 10 at 2:00 p.m. PT. The reason: Anthropic "recently fixed a few small issues that used more limits than intended," and admitted that the resets tied to the Claude Fable launch did little for people who had already hit their weekly cap.
Startup Fortune reports the reset sits under Settings and Usage. Because there is only one, the practical move is to hold it until a weekly limit blocks real work, not burn it on a short session limit that would refresh in a few hours anyway. Free-plan users are not included.
Usage caps are now a selling point on both sides: our breakdown of ChatGPT Pro's usage limits shows how OpenAI's top plan meters the same kind of heavy use.
Adobe MAX 2026: Firefly Builder beta and an MCP for ChatGPT, Claude and Gemini
Adobe used its MAX conference on November 10 to push deeper into agents. The headline for SaaS teams is a local MCP server that exposes features from Photoshop, Illustrator, After Effects and Premiere to outside chatbots such as ChatGPT, Claude and Gemini. Adobe VP Deepa Subramaniam told TechCrunch the company has "no need or desire to sort of limit" what the connection can reach and will expand it based on feedback.
Firefly gets Builder, a prototyping tool in private beta that puts Adobe closer to Figma's territory, plus on-brand charts, editing of one part of an image without touching the rest, automatic video assembly from clips and sound effects, and an always-on agent that gathers context about your projects. Photoshop adds relighting, and Premiere gets new color grading and object removal for video.
Adobe has not published credit costs or general availability dates for these features, so Creative Cloud admins should hold off on budget changes until the generative credit terms appear.
Reddit Reviews: AI summaries of what users say about 10,000 products
Reddit launched Reddit Reviews on November 10 for all US users on iOS, Android and the web. Search for a product, or tap a highlighted product name in a comment, and you get a "what people are saying" summary, a breakdown of positive, negative and mixed comments, and the threads behind it. It starts with about 10,000 products after tests that began in February with electronics, video games and health and beauty, then added home and garden and fashion in September.
For shoppers, it is a faster version of adding "reddit" to a Google search. For brands, including SaaS companies, it means community threads now feed a summary that sits next to the purchase decision. TechCrunch flagged the obvious risk of astroturfing and AI-written comments; Reddit says its site rules, anti-spam systems and volunteer moderators will police it.
Section 2 SaaS and business
Railway breach: $740,000 in fraudulent AI charges, and who pays them
Cloud host Railway now estimates that AI provider keys stolen in its October breach ran up about $740,000 in unauthorized charges, The Hacker News reported on November 11. The attacker escaped build isolation on October 20 and decrypted stored environment variables on October 21, hunting for AI key names. Railway calls 100,000 workspaces an upper bound for exposure; it also reset linked-account tokens for 1,852 users and says no customer databases were accessed.
Railway says it will cover charges that an AI provider refuses to refund, but it has not set a claim deadline or cap. The refund route depends on whose key was abused:
Railway breach: refund route by AI provider
AWS Bedrock
Provider refunds
Existing fraud policy applies; account teams have started refunds
Google Gemini
Provider refunds
Existing fraud policy applies; refunds under way
OpenAI
Case by case
File a claim with OpenAI, then give the case ID to Railway support
Anthropic
Railway covers for now
Talks ongoing; Railway pays charges in the meantime
OpenRouter
Railway covers
No-refund policy stands, so Railway pays all related costs
If you hosted on Railway before October 21: rotate every AI key at the provider, replace any secret stored as a variable, report unexpected bills through Railway's charge form, and check GitHub and container registry audit logs.
Is Jensen Huang leaving Nvidia? What the board is weighing
The Wall Street Journal reported on November 11 that Nvidia's board is preparing to oust co-founder and CEO Jensen Huang, who has run the company for 33 years. Bloomberg added that directors met on Wednesday, have worked with an outside firm to identify replacements and are considering an interim CEO while they search. No decision had been announced when we published.
This is a sharp turn from November 8, when the board opened an independent investigation after Huang admitted an extramarital relationship in a memo to staff and said it still had confidence in him. The New York Times then reported a second relationship, with an executive who reports to him. TechCrunch notes Nvidia stock fell about 8% in after-hours trading on Wednesday and roughly 11% over five days, with no obvious successor in place.
For companies buying AI services, nothing changes in GPU supply today; the story matters because Nvidia's roadmap and pricing set the cost base for most AI tools you pay for.
OpenAI revenue tops $10 billion a quarter as Anthropic's IPO draws heavy demand
OpenAI's quarterly revenue crossed $10 billion for the first time in the third quarter, Bloomberg reported, citing unnamed sources. That is almost 50% more than the $6.7 billion the Wall Street Journal reported for the second quarter. Losses are narrowing but still large: Bloomberg puts the adjusted operating margin at minus 109%, up from minus 150%, and the annual revenue run rate at $50 billion.
OpenAI quarterly revenue, 2026 (reported, $ billions)
Q1
$5.7B
Q2
$6.7B
Q3
$10B+
Anthropic, which says its run rate reached $81 billion at the end of October, is pricing its IPO under the ticker ANT. A day after it set a $70 to $78 range (our November 11 roundup has the share details), Bloomberg reported orders worth multiple times the shares on offer. Filings show it may raise up to $90.4 billion including a concurrent private placement, and a livestream with Dario Amodei and Jamie Dimon is set for Friday. The two companies count annualized revenue differently, so the figures are not a clean comparison.
CoreWeave raises GPU prices again: what it signals for AI tool costs
CoreWeave, the GPU cloud behind a large share of AI startups, reported third-quarter revenue of $3.66 billion on November 10, up 168% from a year earlier, with a net loss of $709 million and a backlog of $132.5 billion. According to TechCrunch, it raised prices about 25% in July and another 10% later in the quarter, and prices are still rising this quarter.
CoreWeave capacity price per megawatt, per year
Short-term deals
$40M+
4–5 year contracts
$20–25M
Short-term capacity now sells for more than $40 million per megawatt a year, against $20 million to $25 million on four- and five-year contracts, and CoreWeave expects about $9 billion in customer prepayments and deposits over the next three months. Our read: when the compute under AI apps costs more and customers pay upfront to secure it, credit-based and usage-based pricing in SaaS tools is more likely to tighten than loosen in 2027 renewals.
FAQ: Cursor, ChatGPT voice, Claude limits and Railway
Does Cursor still have GPT models after November 12?
OpenAI proposed November 12 as the end of its direct model supply to Cursor, though TokenPost says the final date still needs both companies to confirm it. Anthropic and Google models stay available. Some access may continue through your own OpenAI API key or the Codex extension, so check Cursor's model settings before you depend on it.
When does ChatGPT Advanced Voice stop working?
OpenAI retires Advanced and Standard voice on December 11, 2026, and switches users to GPT-Live automatically. Live video and screen sharing stay on Advanced voice until GPT-Live supports them, and your past voice chats remain in your history. OpenAI has not announced any price change tied to the move.
How do I use the free Claude usage reset?
Pro, Max, Ultra and Team subscribers each get one full reset across all models and limits, usable before December 10 at 2:00 p.m. PT. Startup Fortune says it appears under Settings and Usage. Since you only get one, save it for a weekly limit rather than a short session cap.
Will Railway refund fraudulent AI charges from the breach?
Railway says it will cover charges that an AI provider will not refund. AWS Bedrock and Google Gemini are refunding under their fraud policies, OpenAI reviews claims case by case, and Railway pays for Anthropic and OpenRouter charges. Rotate any exposed key with the provider first, because a charge report alone does not revoke it.
Sources: TokenPost (Nov 11, 2026) · Cyber Security News (Aug 29, 2026) · Superframeworks (accessed Nov 12, 2026) · OpenAI Help Center (Nov 10, 2026) · OpenAI Newsroom on X (Nov 10, 2026) · ClaudeDevs on X (Nov 10, 2026) · Startup Fortune (Nov 11, 2026) · TechCrunch on Adobe MAX (Nov 10, 2026) · TechCrunch on Reddit Reviews (Nov 10, 2026) · The Hacker News (Nov 11, 2026) · TechCrunch on Nvidia (Nov 11, 2026) · OfficeChai citing Bloomberg (Nov 12, 2026) · TechCrunch on OpenAI revenue (Nov 11, 2026) · Yahoo Finance citing Bloomberg (Nov 11, 2026) · TechCrunch on CoreWeave (Nov 11, 2026)