OpenAI ships ChatGPT Images 2.6 with GIF output and 6-second image generation
OpenAI just dropped Images 2.6, and it is a solid upgrade worth knowing about.
The headline feature: you can now generate animated GIFs from a text prompt or a photo. You can tweak the speed and edit individual frames. The model keeps characters consistent across frames, so animations actually look coherent.
The other big addition is reference images. Save a character, a pet, or a product once, and reuse it across future creations without re-uploading anything. Huge for anyone building consistent visual content.
On the API side, two models are now available:
- Flare (gpt-image-2.6-flare): speed-first, generates in as little as 6 seconds. Great for prototyping and high-volume pipelines.
- Sunburst (gpt-image-2.6-sunburst): higher quality, more consistent edits. Built for polished output.
Both support image markup and SVG export, so you can pinpoint edits and pipe results directly into design workflows. Overall generation is 50% faster than before. Available now in ChatGPT, Codex, and the API.
Kimi.ai releases Kimi K3.1 Pro model weights to the public
Moonshot AI just dropped the weights for Kimi K3.1 Pro on Hugging Face, and it’s a big deal for anyone who wants a powerful model they can actually run themselves.
So what is it exactly? A 2.8 trillion parameter open-weight model with a 1 million token context window, meaning you can feed it an entire codebase at once. It uses a Mixture-of-Experts setup, where only a slice of the model activates per task, keeping it efficient despite the massive size.
Here’s what it’s good at:
- Full-project coding in Unreal, Unity, and Godot, plus 3D modeling in Blender and CAD
- Finance work with hallucination down by 65%
- Ranked #1 among open-weight models on Agent Arena, across productivity, coding, and chat
Platforms like Ollama, Nebius, and Databricks already support it, so you don’t need to self-host. The weights alone are around 1.56 TB, so local runs need serious hardware.
Musk, Trump and Their AIDC Lost
The two loudest political backers of the AI buildout lost Tuesday’s midterms, and took their data centers down with them. Elon Musk spent at least $474 million, $224 million of it in Texas, on Republicans who lost. Democrats took Congress and the governorships of Texas, Ohio and Georgia, home to about 39% of planned US data center capacity.
Trump says he “wasn’t on the Ballot.” Musk posted: “Data centers in space looking pretty appealing right now.” Neither reply builds anything on the ground. Trump’s 2025 order sped up federal permits, and the freeze coming in Texas is a state one.
Gina Hinojosa, the governor-elect, ran on halting new data centers until the legislature writes rules, and on ending their tax breaks. Jefferies puts about 13 gigawatts of announced Texas projects at risk. SpaceX, which is building data centers in Texas, fell 4%. CoreWeave fell as much as 7%.
JPMorgan prices the AI buildout at about $5 trillion through 2030, roughly $2 trillion of it borrowed. Voters also weighed 41 local data center measures, and at least 10 communities approved bans or restrictions. Trump wasn’t on the ballot. The data centers were.
Green Data
Portugal’s government is finalising measures that will oblige data centres to use renewable energy and water-free cooling technologies, as the country accommodates developments scheduled to provide over 2.6GW in compute capacity.
Slash and Burn
A criminal investigation has been opened into a Google data centre currently under construction in the north of Finland, following allegations that forest clearing and drainage had taken place at the planned site before conducting environmental impact assessments.
Metadata
The UK’s Competition and Markets Authority has opened an investigation into Meta’s new commercial terms, which the tech company has indefinitely postponed in response to widespread criticisms that they would enable it to scrape content from publishers in order to train AI.
OpenAI Ships a Frontier GIF Model
OpenAI has shipped the first frontier image model that animates on its own. Images 2.6, launched Wednesday in ChatGPT, Codex and the API, turns a prompt or a photo into a looping GIF you can speed up and edit frame by frame.
Until now, motion meant video. OpenAI draws the motion itself: its example prompts ask for “a single animation sprite sheet,” played as a loop.
The rest of the update: Images 2.6 is 50% faster, and Flare, the quick API version, can return an image in six seconds. Sunburst, the quality version, holds characters and products steady across edits. Developers can mark the exact spot to change and export SVG files for web and print. Prices did not move.
It still has to prove itself beyond OpenAI’s own demos, and one risk is familiar. Sora, OpenAI’s video app, drew Hollywood’s copyright complaints within a week of launch and was shut down six months later. Memes run on other people’s characters, which is exactly where Sora’s trouble started.
Apple Delivers Siri’s 2024 Promise
At WWDC 2024, Apple said Siri would take more actions in and across apps. More than two years later, it let developers try it in a beta. Apps can ship with it next spring.
Apple calls Siri AI “profoundly more capable.” Its own fine print lists six languages, no EU and no users under 13. The terms and usage policies are still to come: “More to share soon.”
The service is also capped. Server-model features carry daily usage limits, and “expanded access to such features will be available for a fee in the future.”
How good is the model underneath? In June, Apple promised a technical report with “updated evaluations and benchmarks” “later this summer.” The results it did publish compare its new models with its own older ones. The post still says “later this summer.”
Siri took two years. The terms, the table and the price are still coming.
Ram Altman Wants Your RAM
In one interview, Sam Altman said he is “most proud” of “putting these tools in the hands of people.” A good pitch, he said, is “We are going to enable you to have a better life that you define for yourself.”
Later in the same sitting, he called it “quite terrible” for everybody to keep “a bunch of the world’s limited supply of DRAM” in their own computer, used 5% of the time. His fix: “You want maximum utilization in the cloud.”
The internet heard a plan for your RAM. Altman renamed himself Ram Altman on X, said he wants people to “put as much as they want in their computers,” and called the clip taken out of context.
He has run this play before, with ads. In 2024 Altman called ads plus AI “uniquely unsettling” and ads “a last resort,” acceptable only if they were “the only way to get everyone in the world access to great services.” Twenty months later he announced ads in ChatGPT. They are now in 89 countries.
The clip was the context. He wants the tools in your hands and the memory to run them in the cloud.
Atmanity releases a real-time video avatar model
Atmee v2 can turn a single image and 10-second voice clip into lifelike avatars you can talk to. The model took two years to build and comes from founder Chen-Ping Yu, who previously sold his AR startup to Google in 2022 before launching Atmanity. The startup claims its avatars are up to 50x faster and 10-100x cheaper to run than top video models, and can be built by pasting a single link into your preferred chatbot.
OpenAI upgrades its image model to create GIFs
ChatGPT Images 2.6 lets you bring any image to life with subtle motion, a useful update for animating a brand mascot, making product shots more eye-catching, or creating funny GIFs. The new model creates images 50% faster than before and is available now. In separate news, product lead Adam Fry said ChatGPT has reached roughly 20% of Google Search volume. If accurate, it would make the chatbot Google Search’s most serious challenger in decades.
Apple and Goldman Sachs restrict employee AI use for different reasons
Apple is reportedly reducing daily AI limits for some engineers in an effort to cut costs. Meanwhile, Goldman Sachs has limited the use of AI coding tools to its engineering teams, cutting off most traders and bankers over security concerns. It’s a notable reversal considering the bank’s CIO praised vibe coding by non-technical staff in July. Meta and Microsoft made similar moves in early October, steering employees toward in-house tools instead.
How to build an on-demand data analyst for your team with Grok Bot
- Open Grok Bot > click ‘+’ > Create a Bot and create a dedicated Bot for company data analysis.
- Open the Bot’s settings and connect the data sources, files, plugins, and credentials it needs to access your company data.
- Add any company-specific context it should follow, such as metric definitions, table descriptions, reporting rules, and dashboard conventions.
Sample Prompt: “Act as our company data scientist. Investigate [metric/question] using our connected data. Query the relevant tables, check the result by [segment/time period/region], identify the most likely drivers, flag data-quality issues, and return a concise explanation with supporting charts and the exact metrics used.”
- Run several real queries, review its analysis, and update the Bot’s instructions or add a ‘Skill’ for any recurring analysis workflow.
- Open ‘Share’ > ‘Publish to Team’, then choose whether to publish the Bot with its existing context or start the shared version fresh.
- Teammates can then open it from ‘Team Bots’ or use the connected Slack integration to ask data questions on demand.
Google Ties School Search Access to AI
Google’s permission slip has a mandatory attachment.
Google is giving schools a “my way or the highway” choice: disable Search and several AI services by default for managed K-12 accounts in early 2027 or accept AI. If administrators want to keep Google Search available, they must accept its integrated AI features. Schools mandating a 100% AI-free zone will lose student access to Search entirely.
Facing immediate backlash, Google took to social media to clarify that administrators can toggle Search independently from the Gemini chatbots. But the core ultimatum remains intact: there is no magic switch to get traditional search without the generative AI.
Google’s proposed olive branch is Guided Learning, an AI tutor built to coach kids through homework rather than just doing it for them. A Search pilot starts in select US schools this month. But since the tutor is still very much artificial intelligence, it doesn’t exactly solve an AI ban.
Sensing an opening, DuckDuckGo is aggressively pitching itself to frustrated schools. It also offers AI tools, but actually lets users turn them off. Who knew a functional “off” switch would become a killer competitive advantage?
The timing adds some competitive spice: OpenAI says ChatGPT’s search-like activity already equals roughly 20% of Google’s search volume.
For school IT teams, drafting an AI policy now doubles as a search-engine showdown. Google is practically daring its captive education market to test-drive competitors—all because it refuses to give educators the one setting they keep requesting.
Samsung’s Gemini Glasses Eye a December Debut
Samsung wants Gemini to sit on your nose.
Samsung’s attempt to glue Gemini directly to your face could reach the US and South Korea in mid-December, according to Chosun Biz. Since Samsung hasn’t confirmed the date or price—and earlier rumors hyped November—treat this timeline with skepticism. Global availability depends on whether early adopters actually buy the things.
These Android XR frames wear fashionable Gentle Monster and Warby Parker disguises, but don’t expect holograms: they are completely display-free. Instead, a camera, microphones, and speakers let Gemini act as a whispering shoulder angel, delivering spoken translations and directions based on what you see. Your smartphone still runs the show; it just gets to loiter in your pocket.
Samsung also teased a corporate party trick: staring down a meeting whiteboard to capture and automatically organize the scribbles into Samsung Notes. You’ll need compatible Galaxy gear to make the magic happen. The company boasts nine hours of battery life, assuming you aren’t pestering the AI constantly.
It’s a direct shot at Meta’s smart specs. But while fashion partnerships keep you from looking like a cyborg, bringing face-mounted cameras into the office means HR will have a field day. A tiny privacy LED doesn’t write corporate recording policy for you. Perhaps that’s why Meta is already rolling out its own camera-free alternatives.)
Samsung’s pitch: fewer screen checks, more things to charge.
Voters Put Data Centers on a Shorter Leash
US voters backed data center restrictions and rejected development proposals in dozens of local ballot decisions on Nov. 3, proving that ordering a mountain of GPUs doesn’t automatically come with a neighborhood permission slip.
The measures span outright bans, power-use caps, zoning decisions, and nonbinding advisory questions. In Trenton, Ohio, nearly 80% supported banning data centers above 25 megawatts—because apparently, 24 megawatts is the exact limit of local patience. Advisory votes elsewhere register fierce opposition without automatically halting construction.
Overall, the objections are concrete. Residents are tired of footing the bill for soaring utility costs while listening to the sweet, sweet hum of a million server fans. The anti-AI sentiment even reached the statehouses, as Texas governor-elect Gina Hinojosa and Ohio governor-elect Amy Acton successfully campaigned for moratoriums. Of course, those pledges still need to survive the leap into actual policy.
One awkward possibility is that these grassroots restrictions could just strengthen established cloud giants. Stratechery’s Ben Thompson argues that tech titans with existing sites and permits may gain a massive advantage over smaller providers desperately seeking new locations. Making construction harder doesn’t necessarily make Big Tech smaller—it just pulls the ladder up behind them.
The eventual hit to total capacity remains uncertain. For companies banking on the AI boom, though, the results inject immediate political turbulence into deployment timelines. Developers now face a wildly difficult new hurdle: actually having to convince local communities that the disruption is worth it or, in SpaceX’s case, offering to buy over 400 homes just to quiet a turbine noise lawsuit.
The cloud has discovered a very terrestrial dependency: permission.