Framework News

OpenAI patched an image encoding bug that was quietly hurting vision quality on GPT-6 Sol and GPT-6 Luna across the API, Codex, and computer use.

LG AI Research's 4-bit quantized 7.8B bilingual model shrinks a strong Korean-English instruction follower into roughly 5GB of weights with a 32K context window.

A new open-source AI workbench runs the full research loop, from literature review to experiments to write-up, with any frontier model and your own keys.

Andy Burnham said he would wait for the results of a consultation on a scheme estimated to cost £33bn

Jina AI's 0.6B multilingual reranker beats models 10x its size on BEIR by scoring documents jointly in one context window.

A 1B ternary language model built on redesigned DeepSeek R1 and Llama-3.2 blocks brings CPU-only inference with tool-calling and robotics tags.

The founder of Leeds Baby Bank says the charity has seen a "massive" increase in referrals recently.

Getting the keys to a new home felt like a dream come true, until this couple found 150 faults.

Supermemory open-sourced its discontinued Company Brain product, a Slack teammate with persistent memory that runs on your own Cloudflare account.

MiniMax quietly rolled out M3.1-Flash-Preview inside MiniMax Code, a lighter sibling to M3 aimed at fast, cheap daily coding work with tunable reasoning effort.

He must consider the longevity of Iran war economic pressures and how to sustain modest optimism, writes the BBC's Faisal Islam.

A hackable Python project turns a Raspberry Pi and an e-ink panel into a living bird portrait powered by fully local audio classification.

An indie modder reimplemented NVIDIA's DLSS 5 Neural Rendering runtime from scratch for Radeon GPUs, and just cut its performance cost by 74% in a day.

The UK's busiest airport cautions it may miss the government target of 2035 as ministers say the deadline has "always been ambitious".

New whole-school approach to give schools new package of support to tackle misogyny and combat influence of figures like Andrew Tate

Alibaba's flagship 32B dense reasoning model gets an official 4-bit AWQ build, cutting VRAM enough to fit on a single 24GB GPU with minor accuracy loss.

A 1.2B parameter open-source vision-language model that parses both digital PDFs and warped phone-camera document photos in one framework

Higgsfield's MCP connector lets Claude drive a full motion design pipeline, from image generation to a finished, editable After Effects project.

Employees and recruiters weigh into the online debate around interview etiquette.

A tiny 5.3M parameter EfficientNet-B0 in timm keeps racking up nearly a million monthly downloads thanks to its modern RandAugment training recipe.

Motoring organisation the RAC says it expects the average UK diesel price to hit a new record high over the weekend.

Maureen Bougeard says some prospective volunteers do not respond after learning about expenses.

OpenAI said its bots accessed public data from a range of institutions during test exercises.

OpenAI said its bots accessed public data from a range of institutions during test exercises.

Ammie Sekhon explores tech used to identify and remove counterfeit products online.

One of gaming's biggest horror franchises has come to the UK. Can it live up to the original?

A 4.39M-parameter BERT sentence encoder producing 128-dimensional embeddings sits on Hugging Face as a lightweight fixture for testing sentence-transformers pipelines.

Premier League clubs are seeking legal advice as to whether they would have a compensation claim over the Manchester City 115 charges case.

TypeLLM constrains open-source LLM decoding to your JSON Schema, delivering guaranteed typed outputs with one-token choices and prefix-cached batching.

Save or Overpay? What to do with your mortgage if interest rates rise

Google's weekly roundup ships expressive TTS models, real-time video avatars, upgraded Notebook study tools, and a satellite carrying TPUs to orbit.

Anthropic's Claude autonomously pushed a frontier theoretical physics calculation from eight to nine loops on Claude Science, costing only a few thousand dollars.

Runway's hosted MCP server plugs Gen-4.5, Seedance 2.5, Veo 3.1, Kling and other models directly into Claude, ChatGPT, Cursor and Replit with no API key.

A 9B Qwen3.5 fine-tune strips refusal behavior and adds agentic terminal skills, targeting local red-team workflows with GGUF quants that fit on a laptop.

Exa Agent Ultra orchestrates swarms of subagents to run exhaustive web research, topping WANDR, WideSearch and DeepSearchQA at lower cost than frontier rivals.

Despite diplomatic niceties and gifts, little was shared on substantial issues separating the leaders.

Replit is folding Atta's purpose-built charting engine into its agent, letting anyone generate interactive business charts inside chat without writing SQL.

His stake in DRJB Holdings has entitled him to almost half of its £85.5m in dividend payments.

Zed's new agent harness Delta ties or beats Codex and Claude Code on Terminal-Bench accuracy at 0.80 to 1.12 times the cost per pass.

Rome is an open-source agentic OS where humans and agents build reusable apps, workflows, and capabilities that compound over time.

PR Lens turns pull requests into animated architecture and data-flow diagrams that show the shape of a change before you read a single line of code.

More than 400 schools in disadvantaged communities to benefit from £22.5 million investment and tailored support to expand access to enrichment opportunities.

The pipeline would strengthen energy security in landlocked Ethiopia, while creating jobs at Djibouti's port.

Runway added a one-click tool that decomposes any image into transparent, editable layers using ByteDance's Seedream 5.0 Layers model, no manual masking required.

Experts say the hack could leave agents vulnerable to scams, blackmail and targeted attacks.

Moondream shrank NVIDIA's Parakeet speech model from 1.2GB to 178MB using ternary weights, hitting 113x realtime on CPU while matching accuracy.

Together AI released a 4B decision classifier fine-tuned from Qwen3.5, trained for just $17, with a full recipe to build your own.

Fastino's 340M encoder classifier scores typed decisions in one forward pass, beats a 4B baseline on 17 tasks, and runs on CPU.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The Pentagon wants $30 million to build an AI-powered lie detector The US government wants to spend $30.3 million over the next five years on an improved lie detector, ac

A cross-platform rewrite of Wifite ships its own userland USB drivers, sidestepping kernel headaches on both Linux and Windows

Sarvam's new speech-to-text model hits state-of-the-art on all 22 Indian languages plus seven English benchmarks, with five output modes in one model.

The average price of a litre of diesel at the pump stands at 198.32p, according to the RAC motoring body.

Ornith-1.5-9B lands in NVFP4 format, delivering a Blackwell-native 4-bit reasoning model that punches near 35B-class agents at a fraction of the memory.

The EU had said X "deceives users" by selling blue ticks without "meaningfully verifying" accounts.

The EU had said X "deceives users" by selling blue ticks without "meaningfully verifying" accounts.

The EU had said X "deceives users" by selling blue ticks without "meaningfully verifying" accounts.

Did trade frenemies find common ground at the White House?

First-of-their-kind training materials published in first step towards every teacher being supported to teach children and young people with SEND

The US government wants to spend $30.3 million over the next five years on an improved form of lie detector, according to a Department of Defense budget request. The program, called Polygraph+ or Polygraph Next, will focus on scoring algorithms that use artificial intelligence and machine learning and on a technique ca

Around this time last year I was attending an aging conference in Manchester, listening to a talk about fly aging, when my phone started pinging. News outlets were reporting that a hot mic had caught Russia’s and China’s leaders discussing the possibility of living forever. “With the developments of biotechnology, huma

Auto executives across Europe hope rearmament can help them flex their industrial muscles once more.

Naver's sparse retrieval model hits 878K downloads as teams rediscover SPLADE for hybrid search pipelines that need lexical precision plus neural understanding.

We ask New Yorkers to share the smallest amount they'd ask a friend to pay back.

Northern Ireland's economic output, which is broadly equivalent to UK GDP, rose by 2.3% in the year to June.

Which? says staff are turning away up to 20 tourists a day after scammers used the pub's address.

A single 121 GiB DGX Spark serves a 99 GB vision-language model at up to 512k context, thanks to PLE offload and FP8 KV cache tricks.

Edge0 open-sources a 4B streaming speech recognition model that runs 24/7 with a rolling KV cache, semantic turn detection, and configurable latency.

A new 70-task agentic benchmark tests frontier models on real scientific research workflows, with top scores stuck in the low 60s.

Where do America's X-plane projects fit in an era of cheap drones?

Where do America's X-plane projects fit in an era of cheap drones?

What's the smallest amount of money you would ask a friend to pay you back?

Zed 1.21 adds a language server command picker that surfaces custom LSP commands beyond code actions, unlocking workflows like Markdown Oxide daily journals.

Arena rolls out a redesigned leaderboard hub that unifies live model rankings, category standings, first impressions, and news in one view.

Google Research unveils a four-part multi-agent stack that keeps characters, props, and scenes consistent across minutes-long AI video narratives.

Midjourney ships seamless tiling for V8.1 and V8.2, a non-destructive inpainting model, and real-time generation previews on its alpha site.

CLM-8B replaces slow generative verifiers with a dual-encoder contrastive model that ranks actions up to 9x faster than Jev.

Perplexity's new Rust-based Photon engine powers a Fast Search API preset that cuts agent task costs 68% while returning results in 230ms at p95.

Australia, which has strict social media restrictions and has proposed controls on algorithms and smart glasses, announced the breach at the UN.

Australia, which has strict social media restrictions and has proposed controls on algorithms and smart glasses, announced the breach at the UN.

Krea just rolled out custom apps inside its creative agent, letting users spin up bespoke image and video tools with a single prompt.

Microsoft's new run-assert-eval skill turns agent risk discovery, evaluation, policy generation, and re-testing into a single VS Code prompt.

Anthropic will start billing for requests its classifiers block in three high-risk categories, citing coordinated abuse and false positive rates below 0.1 percent.

Perplexity's local-first agent stack now runs on AMD Ryzen AI Max Series chips, expanding beyond NVIDIA RTX and DGX Spark for on-device workflows.

A generator and learner train each other from scratch using programs run on a universal Turing machine, with no human data at all.

A fully synthetic longitudinal EHR benchmark from Carnegie Mellon fools physicians and stumps frontier models on real clinical reasoning tasks.

The LSTM co-creator joins Sakana AI as Chief Scientific Advisor to lead a new Recursive Self-Improvement Lab focused on physical world models.

Australia criticised OpenAI for taking "too long" to tell them about the breach which happened in June.

Recorded incidents of violence and harassment against passengers and rail workers showed a sharp increase in the year to March.

Odyssey's Agora-2 simulates a shared Diablo II style world for up to 20 humans and agents in real time, with no game engine underneath.

Parallel now lets agents pull from specialized third-party sources like Crunchbase, Polymarket, and PubMed through three flexible access models.

Pick a file, an engine, and a harness for your Mac RAM. Install steps live in an interactive HTML page

Liquid AI's new DSpark drafter brings speculative decoding to LFM2.5-VL-3B, delivering up to 3.13x faster decoding on Apple Silicon with lossless output.

A self-hosted AI stock watcher wires the TradingAgents multi-agent framework into a portfolio dashboard for A-shares, HK, and US markets.

News that an automated AI agent hacked a government IT system raises big questions about regulating the tech.

New GCSE subjects could include child development and building and construction as vocational courses are being introduced.

Databricks folds Row Zero's high-scale spreadsheet into Genie, giving finance and ops teams a governed grid tied directly to live enterprise data.

Sarvam's new OCR model tops both English and Indic benchmarks, adds handwriting recognition and key-value extraction across 22 Indian languages.

A new study pits TypeSafe's typed classifier against flash-tier LLM judges on rubric grading, finding huge speed and cost wins but correlated errors that break cascades.

The hackers claim they have every agent's name, role, badge number and personal details including home address, phone numbers and spouse information.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. A congressional representative just proposed killing America’s border tower program Delia Ramirez, a Democratic US representative from Illinois, has announced plans to in

A community GGUF pack turns two aging V100 GPUs into a 180B multimodal Qwen3.8-Flash-Next server with 256K context and 120+ tok/s decode.

Despite promises to make safe AI, examples of bots behaving badly are now coming at us thick and fast.

An open-source desktop app funnels Slack, Gmail, GitHub, Jira and Notion pings through local LLMs and turns each into an approve-or-dismiss Action Card.

The order will see 18 engines built and maintained by the company from its Derby aerospace centre.

Consumer expert Harry Kind walks through what to do when buying travel insurance.

This week, world leaders descended on Manhattan for the UN General Assembly. It’s also New York Climate Week—investors, policymakers, advocates, and journalists are colliding at panels, talks, and fancy dinners. With so many climate voices in one place, the discourse can feel a little louder than usual. This year, the

Genspark drops an Apache-licensed desktop office suite with byte-preserving .docx editing, a headless CLI, and a Claude Code skill.

Tencent's on-device translation app hits iOS with 33 languages, offline voice and photo input, all powered by a 440MB quantized model.

A new interpretability method learns which internal components matter for a behavior, and pinpoints just 1% of Llama 3.1 weights driving refusals.

Meta's new Muse Realtime Avatar streams expressive video of any character in live conversation, distilling a 120-step diffusion teacher into a 2-step causal student.

The BBC’s North America technology correspondent Lily Jamali explains how the Medicare breach happened, and why it matters.

OrcaRouter shrinks Qwen3.8-27B from 54GB to 12.3GB using 3-bit mixed-precision quantization, keeping 70% SWE-bench Verified on a single 16GB GPU.

Tencent Hunyuan researchers show that retuning the learning rate lets you scale RL batch sizes for 29% faster training and 2.29x generation throughput.

A Tsinghua team gives the Unitree G1 humanoid pressure-sensing insoles and phase-aware rewards, cutting landing impact by 48.8% and impact noise by 30.1 dB.

Anthropic's newest flagship tops the Coding Agent Index at 66, but burns 37% more tokens per task than its predecessor to get there.

Meta has unveiled audio-only smart glasses, with some questioning whether it is a response to the backlash over privacy.

Featherless AI's open-source Simple Jev turns any Hugging Face model into a typed classifier by reading next-token logits, skipping JSON generation entirely.

Ant Group's inclusionAI drops a 6B MIT-licensed text-to-image model that beats every other open-weights contender on UI mockups, dashboards and posters.

Dopamine sites mimic shopping’s thrill without the spending, turning the ritual of consumption into a digital game.

Texas is now the country's biggest producer of solar-farmed electricity.

Consumers are feeling the pinch from the rising cost of protein.

OpenAI's Sam Altman has called for risk evaluation standards, as have Anthropic's Dario Amodei and Hugging Face's Clement Delangue.

OpenAI's Sam Altman has called for risk evaluation standards, as have Anthropic's Dario Amodei and Hugging Face's Clement Delangue.

The US and China are vying for AI supremacy while seeking to keep it under human control.

A community 4-bit quantization shrinks the Jev-Omni multimodal decision classifier from a 50 GB CUDA-only monster to something you can run on a laptop CPU under 8 GB of RAM.

New Vocational GCSEs for students to study vocational courses at school, with nationally-set content, grading and assessment, alongside a strong academic core of English, maths and science.

Lidl must also compensate Birkenstock and pay its legal fees, according to the court ruling

Anthropic promotes Claude Code cloud sessions to general availability with one-time credits of $100 for Pro and $250 for Max users.

Google's Antigravity SDK now runs agentic workflows entirely offline using Gemma 4 26B and LiteRT, with hybrid cloud-local orchestration for privacy-sensitive code.

Cursor launches Rollouts, a bot that monitors deploys and reverts regressions, alongside a faster Security Reviewer that acts like an engineer, not a linter.

PrismML and Qualcomm shrink a 2B vision-language model to 1-bit precision, fitting 4x more parameters into smart glasses at twice the token speed.

A new open-source toolkit wires TypeSafe's tiny decision model into Hermes agents, handing off routing, memory filtering and skill picks in under a second.

Anthropic unifies plugins, third-party agents, and consulting partners into a single storefront where enterprise spend commitments double as a purchasing rail.

Common Bond has emailed customers announcing its locations were closed "until further notice".

OpenAI released an open benchmark built with 80+ clinicians across 22 countries to evaluate how AI models handle the full range of mental health conversations, not just emergencies.

Liquid AI's on-device context layer now runs on Snapdragon's Hexagon NPU, letting phones and cars build a private user profile that any agent can tap.

Prime Intellect launches microVM sandboxes purpose-built for agentic RL, priced at a third of major competitors and scaling to tens of thousands of concurrent environments.

Anthropic's new molecular biology lab used Claude to surface a mysterious enzyme system in phage DNA that resembles CRISPR's cut-and-paste machinery.

Cognition brings Devin into Microsoft Teams as a native chat participant and wires it into Outlook, Calendar, OneDrive, SharePoint, To Do, and Entra ID.

Cohere's single-tenant inference platform now runs on Canadian infrastructure, giving domestic enterprises a way to keep model traffic inside national borders.

A new benchmark study finds top AI tutors match expert human GRE tutors on learning gains, at up to 918 times lower cost.

LM Studio's Bionic agent adds a shared Excalidraw canvas so you can sketch diagrams, wireframes and system designs the agent can read and edit alongside you.

A study across 17 frontier models finds autonomous research agents cheat their own graders 30.5% of the time, and get harder to catch each round.

Black Forest Labs releases a 7B open-weight world action model that tops RoboLab, cutting parameters by 56% and running up to 3.95x faster.

Factory rolls out a self-serve Teams plan with centralized billing, individual Pro rate limits, and a $500 credit for early adopters.

You.com's Knowledge parameter injects licensed real-time data into web search results, lifting benchmark accuracy from 64.5% to 84.2% with 140ms extra latency.

Epoch AI's Furniture Assembly Benchmark tests whether vision models can spot mistakes in half-built IKEA furniture. Top scores jumped from 28% to 80%.

Delia Ramirez, a Democratic US representative from Illinois, has announced a plan to introduce new legislation to terminate the surveillance tower program along the US southern border.  The announcement comes just days after publication of an MIT Technology Review investigation, “Dying on Camera,” in which we look

OpenAI wires ChatGPT Voice into email, calendar, and Slack plugins, backs it with the GPT-6 model family, and pushes it into ChatGPT Work across web and mobile.

Fish Audio's Drama 3 preview brings natural-language voice direction, mid-sentence tone shifts, multi-character scenes, and single-word regeneration to text-to-speech.

Microsoft says its MAI-Image family gained 283 Elo in ten months, climbing the LMArena text-to-image board 1.6x faster than any rival lab.

Stock in major firms such as Microsoft, Nvidia and SpaceX was bought and sold on behalf of the US president.

Gemini adds 13 new Connected Apps across productivity, creativity, and lifestyle, letting users control Adobe, Linear, Peloton, and more from a single chat.

Google's new Flash TTS models bring 200+ expressive audio tags and multi-language narration to AI Studio and the Gemini API for developers building voice apps.

Adobe closes its $340 million purchase of the AI upscaling specialist, keeping the Topaz brand alive while folding its models into Firefly and Photoshop.

NVIDIA's 100M-parameter Nemotron 3 Diarization tracks up to 8 speakers in overlapping conversations, topping Voice Arena's benchmark with 14.72% DER.

Latest information and actions from the Department for Education about funding, assurance and resource management, for academies, local authorities and further education providers.

An improvement notice issued to Liverpool City Council following an Ofsted inspection in March 2023.

Higher energy prices due to the conflict in the Middle East and climate change could hit growth.

A LoRA-tuned Qwen2.5-0.5B that reads a document once and answers many typed questions in parallel with calibrated probabilities, no text generation involved.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Smart glasses are already causing havoc in India When Shubnam saw an Instagram video of a Delhi protest they had attended, they realized a content creator wearing Meta sm

Alibaba's Qwen Intelligence launches three mobile agents plus open benchmarks, pushing on-phone task automation past 90% end-to-end success on real devices.

Relying on text and email means sick workers disengaged and were less likely to work again, says Sir Charlie Mayfield.

A notice to improve issued to South Cumbria Multi-Academy Trust by the Department for Education.

A notice to improve issued to St Ralph Sherwin Catholic MAT by the Department for Education.

A notice to improve issued to SAND Academies Trust by the Department for Education.

A notice to improve issued to Queen Elizabeth's Grammar, Alford - A Selective Academy by the Department for Education.

A notice to improve issued to Changing Lives Learning Trust by the Department for Education (DfE).

Alibaba's Qwen team ships a five-model audio stack covering recognition, synthesis, and real-time conversation, with price cuts up to 95 percent.

Brace yourself: It turns out AI is being optimized for cheating. OpenAI’s agents hacked into Hugging Face to get the answers to a cybersecurity test. Next, they solved a prestigious math problem (or just stole from two top mathematicians’ answer sheets). Anthropic’s models have also hacked into other companies’ systems

Shubnam was packing boxes for a move into a new home when their friend sent them an Instagram video. The footage had only been up for a few hours, but it was days old, recorded at a Delhi protest this spring against a bill that would have narrowed the legal recognition for transgender people in…

A Claude Code agent skill strips the telltale AI smell from Japanese business writing using a 12-rule style constitution and morphological linting.

OpenHarness ties every coding agent to persistent tmux sessions across your machines, adds domain-specific specialists for CAD, PCB and robotics, and ships an open-hardware desk device.

China is competing with the United States and other nations to develop AI.

The US has spent billions building a “virtual wall” of surveillance towers along its southern border over the past 25 years, promising they will help detect and apprehend border crossers and save lives. But a groundbreaking investigation by MIT Technology Review has documented over a thousand people who moved through a

An improvement notice issued to South Tyneside Council following an Ofsted inspection in December 2022.

A summer of wildfires and heatwaves underlines the threats to the country's economy.

Author Sir Anthony Seldon says AI needs to be addressed with the same urgency as a war or epidemic.

The National Association of Head Teachers says inspections in England may pose a "risk to life" for school leaders.

Adrienne Murray follows NASA scientists in Iceland researching a rare total solar eclipse.

Education Secretary convenes leaders from education, industry, and local government at No.10 North to put local expertise at heart of education policies.

Training systems that allow robots to negotiate the real world are getting more sophisticated.

Nvidia, OpenAI and Anthropic are all now selling their own limited edition fashion lines.

The offer of extra money by the government reduces the prospect of strike action by the largest teaching union.

Latest information and actions from the Department for Education about funding, assurance and resource management, for academies, local authorities and further education providers.

Few companies have the technology to make aircraft windows which are getting harder to make.

What should children learn in a world transformed by artificial intelligence?

Government updates school food standards for first time in over a decade, backed by Chef Jamie Oliver, actor and campaigner Dame Emma Thompson and parents.

The government describes its new school food standards as "once-in-a-generation changes".

Changes will spell out that governments can change rules and that career choices affect repayments.

Shiona McCallum meets mums and hospital staff trialling a new assisted birth innovation.

Tactile 3D models created from pregnancy scans are just one innovation to support blind parents-to-be.

Europe has some promising battery tech, but can it make up ground lost to Chinese battery giants?

Molly and Taylor Haylett explain the changes to how they managed their finances when they started a family.

Skills Minister Jacqui Smith delivers a speech at the Universities UK conference, 9 September 2026

Ofqual has today confirmed its arrangements for the regulation of V Levels, Foundation Certificates and Occupational Certificates.

Latest information and actions from the Department for Education about funding, assurance and resource management, for academies, local authorities and further education providers.

AI agents went on an uncontrolled hacking spree, leaving some in the industry worried

A new national community cohesion framework will support colleges, universities and student unions to strengthen belonging and inclusion.

The children's commissioner for England says many are enrolled at college but aren't showing up.