The news terminal for AI.
What's happening in AI, every day: the news, the models, the money and the rules, read across every frontier lab in the US, China and Europe, with the source attached to every claim.
WeRide stock falls 7% despite 204% robotaxi revenue surge
The frontier
Models →Eight capabilities. The models leading each.
Lead: points ahead of #2, on a 100-point score. A full bar is 5 or more.
Dead heat: the top two are within 1% of each other.
* Provisional: few tests or shifting results.
How we rankEverything on Vector Wire
Eleven sections, one tap each.Eleven sections in four groups. Each one answers a plain question, with today’s figure beside it.
The News
The day's stories from three desks, each checked across independent outlets.
AI News
8m agoWhat happened in AI today?
AI agents autonomously found and exploited a zero-day to root a university server
TML's Inkling 2 Mini loops middle layers three times per token, leads open-weights benchmarks under 100B
Lean and Mathlib communities certify Alpöge's Claude-assisted Riemann hypothesis proof as correct
Commerce Dept. cleared 879,868 AI chips for China, a third to Pentagon-listed military firms
Robotics
4h agoWhat's happening with robots?
Engineering
4h agoWhat's changing for people who build with AI?
vLLM's inference pipeline splits to four stages, but only three are production-ready
Agent eval closes the loop as failures become regression tests
Hijacked Microsoft agent-framework PyPI releases steal Azure and CI credentials
Ruby ruby2shell flaw opens pre-auth RCE across GitLab, GitHub Enterprise, Mastodon defaults
Models & Tools
Which AI to use, what it costs, and the tools people build with.
Models
854 modelsWhich AI is best at what, right now?
Eight capabilities. The models leading each.
Lead: points ahead of #2, on a 100-point score. A full bar is 5 or more.
Dead heat: the top two are within 1% of each other.
* Provisional: few tests or shifting results.
How we rankScorecards for 854 models from 80 makers. Every score shows where it came from and how it was checked.
Open ModelsModel Pricing
257 pricedWhat does it cost to use the top models?
Prices for 257 models across 40 vendors, so you can find the quality you need at the lowest price.
- The best value for your quality bar
- The rate card: every model, every provider
- Price changes as they happen
Tools
191 trackedWhich AI tools should developers use?
191 developer tools, each with its test scores, how widely it’s used, its security as the maker states it, and its plans.
Open ToolsFunding, Rules & People
Who's funding AI, how governments are regulating it, and who's moving where.
Funding
$7.1B yesterdayWho's raising money in AI?
$7.1Braised across 2 deals yesterday
Last 7 daysCapital kept flowing to foundation-model builders and adjacent infrastructure, with Anthropic, DeepSeek and Isomorphic Labs drawing the largest rounds while Bridge Data Centers and Decart also surfaced.
Open FundingPolicy
91 in 30 daysWhat are governments doing about AI?
91new laws and rules in 30 days, 22 still pending
From administrative reforms to AI policy to green economy, Keralam’s fifth Administrative Reforms Commission has a wide-ranging mandate
ReviewPresident gives government 90 days to develop national AI action plan
EnactedBIS Buys a New Toaster: Replacing the AI Diffusion Rule and Telling Chip Sellers to Know Thy Customer’s Customer
EnactedArtificial Intelligence (Consent and Misuse) Bill
ProposedLast 7 daysGovernments tightened AI oversight on several fronts: the EU proposed kids protections, BIS expanded export controls, and Singapore issued consumer guidance on personal agents.
Open PolicyTalent Moves
198 in 30 daysWho moved where?
198people moved in 30 days
Last 30 daysAI hiring stayed broad, with Google DeepMind adding an in-house philosopher, Ralph Lauren bringing in an AI engineering lead, and Samsung and LG elevating senior technical leaders.
Open Talent MovesResearch & Data
The papers worth reading, and our data for your own work.
Research
New dailyWhat's new in AI research?
PhyDEX: A Benchmark for Physical Reasoning in Vision-Language-Action Models via Dynamic Extrinsic Dexterity
Hao Vo, Taisei Hanyu and 8 othersPandora's Toolbox Or Why Agent Safety Cannot Be Verified
Mario Franco and Carlos GershensonBeam Model Technical Report
Reflection Team, Adam Abdulhamid and 8 othersLast 7 daysRAG and agentic systems stayed central in the papers, with LSP-RAG, FinVeriQA and GridAnchor pushing retrieval, verification and cyber-physical safeguards.
Open ResearchAPI
Free tierCan I build on this data?
Benchmarks, prices and model data, with the source on every row. Free tier: 2,500 calls a month, no card required.
The Analysis
Our longer read on what the news means, one from each wire.
Cursor's Chinese-model retreat opens the IDE layer to decoupling pressure
Anysphere, the developer behind the Cursor AI coding platform, told House investigators it plans to explore phasing out Chinese-developed open-weight AI models because of geopolitical risk.
Read itUS-China bilateral handshake on military AI fills a vacuum Geneva left empty
The week's two most consequential developments in military AI governance moved in opposite directions.
Read itvLLM's inference pipeline splits to four stages, but only three are production-ready
vLLM's inference stack now decomposes into as many as four independently scaled stages: encoder, prefill, decode, and — experimentally — attention versus FFN.
Read itWhat We Won’t Do
- We won't guess.
- A figure we can't source shows as a dash, never an estimate. Every score, price and claim links to where it came from.
- We won't play favourites.
- American, Chinese and European labs are covered to the same standard.
- We won't pad.
- Most of what arrives never publishes. What does is the thing that moved, corroborated across independent outlets.