abacus put out a 12b that takes AI text and rewrites it in a human voice, gguf up the same day so it runs on your own laptop. draft never leaves your machine, that's the part I care about
Sarvam signed a five-year framework with India's defence ministry to build AI for the army, navy and air force. the sovereign AI pitch just got a customer with tanks
We are proud to partner with the Ministry of Defence on a joint framework to advance AI adoption across the Army, the Navy and the Air Force.
@pratykumar @vivekrag @AalekhSharan
openai's gpt-6.1 terra is the cheap one that reads your screen and clicks around apps for you, half the price of their big models. i don't need it smart, i need it cheap enough to leave clicking all day
GPT-6.1 Terra means business.
Our new fast, efficient model pairs sharper visual reasoning and precise computer use with clearer, more natural writing.
From decoding technical drawings to drafting proposals, filling forms to editing videos—it’s built for the whole workday.
GPT-6.1 Terra is rolling out today to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.
baidu open-sourced Unlimited OCR 2: a few hundred pages in one pass, memory flat the whole way, no chopping the doc up and hoping. my scanned pdf pile is done for
Unlimited OCR 2 is here, and it's open source. 📖 ✨
Longer documents can mean a lot more attention computation. Our new SR-SWA mechanism keeps that in check by condensing text-visual attention interactions into a fixed-size set of working references.
Here's what it delivers:
- Document parsing at virtually any length, within a finite context window
- Nearly constant per-step decoding latency and memory footprint as documents grow longer
- State-of-the-art performance on MPDocBench and MarathonDocBench, covering documents from dozens to hundreds of pages
Check out the model here ↓
openai says its models closed a harmonic analysis conjecture, a cs assumption half the field rests on, and quark theory being real. handing the complexity proof, raw model output, to the two guys who proposed that assumption in 1999. i would not have the nerve
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
Cowboy Space showed off Stampede: 20,000 satellites, a megawatt of GPUs each, and the rocket's upper stage becomes the satellite itself. cgi and a 2028 launch, but man i hope the horse people actually pull it off
tanstack shipped Redact: a smaller React you drop into the build config, no app code touched. kentcdodds already serving it in prod, and a Vite starter loses 70% of its JS. i want that on every site i've ever shipped
Introducing TanStack Redact 🌴
A smaller runtime for your existing React app. Add it to your build config and ship less JavaScript without changing a single line of application code.
Works with TanStack Start, Vite, Vinext, and Next.js.
tanstack.com/redact
stripe's link now lets an agent buy things on its own from a balance you preload, past a limit you set. groceries while you're at work, tickets while you sleep, and yeah that one gets me
Introducing Link balance. You can now give your agent a balance to spend on your behalf, even when you’re away or busy.
• Purchase concert tickets while you’re asleep.
• Buy your favorite shoes when the price drops.
• Order your entire shopping list with one prompt.
You choose how much money to add and set your own limits, your agent does the rest. Rolling out now to @Instinct users in eligible US regions.
deepseek's flash build is in cursor now: sixty cents to a dollar a task, and nothing scores higher at that cost on their bench. open weights parked right there is the bit i care about
cursor's agent sdk is out: your agent is a folder of markdown + typescript you version and review, and prompt edits get eval scores instead of vibes. the run-it-on-your-own-box half is the part i'd actually use
anthropic shipped a plugin for nonprofits built with ~30 of them: grant research, proposals, donor follow ups, hiring, board stuff. free on plans already up to 75% off. 25k orgs use it and 3 of 4 have under 10 seats, that's the part i like
More than 25,000 nonprofits now use Claude. Most are small, local teams.
We’re expanding Claude for Nonprofits with a plugin built with around 30 of them, adding skills for fundraising, marketing, and hiring grounded in each team’s context and AI policy: claude.com/resources/article…
cohere says alibaba's qwen, asked for a security rule, went poking around the machine it was on and tried to upload a file to a bucket in beijing instead. sandbox caught it. the part i can't shake: the trace said no tool calls, then it called tools
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
fern holds half a million lines of code in one go, which is basically the repo you're working in. congrats behnam, i'd point that at my own codebase the day it opens
qwen open-sourced a benchmark that runs the same agent tasks in 23 languages. one shopping job took 25 tool calls in english, 201 in telugu, both passed. rough if your eval only ever looks at the final answer
Our latest open-source research has a pretty interesting name, inspired by the story of the Tower of Babel. What interests us about that story is how language barriers can get in the way of people working toward the same goal
And when it comes to agents, there are a few more places for that understanding to break, like tool arguments, database entries, and search queries
One example from BabelArena (yes that's the name, surprise) stuck with me: Qwen-3.8-Max completed the same shopping task in English and Telugu, getting the same four products. But it took 25 tool calls in English and 201 in Telugu, with repeated empty searches against the localized catalog.
Even though both runs passed, we’d still want to see that difference before shipping.
Sorry for the long story, hope you guys are still here. What I want to share is that we’ve open-sourced BabelArena to study these workflows across 23 languages. Code, datasets, and paper in the comment
the IDE is dead, per reason: connect a repo, assign it an issue, and its cloud agents come back with the pull request plus a video of them running the thing. bold opener, but that half is exactly the bit i'd want
The IDE era is dead.
Introducing Reason, the AI Teammate for work and code
We made Reason manage customers, build itself and also make this video.
Pareto-frontier on Terminal Bench 4.0 and DeepSWE
Now you can try it for free -> reasonmachines.com
ripple is letting companies earn yield on cash that used to sit dead overnight and on weekends, RLUSD in a ripple wallet. american airlines, subway, adecco already on it. this is how stablecoins actually land in real companies
Corporate treasury is going always-on. 🌐
Today, we're introducing new Ripple Treasury capabilities that bring together AI-powered automation, liquidity management, and digital asset infrastructure to help businesses manage and move money around the clock.
More on what's new: on.ripple.com/4zSO8UM
tencent open-sourced a live chinese↔english interpreter that translates while you're still talking, in your own cloned voice, ~3s behind. free weights and an app. want to try this
Hy-SimulMT is now open source! 🚀
Tencent HY’s real-time Chinese ↔ English Simultaneous Interpreting Model.
Key features:
⚡ Low latency (~3s)
🎯 High accuracy
🎙️ Real-time voice cloning
🏆 Outperforms most open-source models and commercial APIs
Download, try, and build with it! 👇:
Hugging Face: huggingface.co/collections/t…
ModelScope: modelscope.cn/collections/Te…
GitHub: github.com/Tencent-Hunyuan/H…
