Here’s how this unified-memory superchip will upend the CPU wars, rescue Windows on Arm gaming, and set an expensive new bar for local AI. https://t.co/Z7OPqWNxPe
DeepSeek just made its 75% price cut on V4-Pro permanent. Xiaomi's MiMo slashed V2.5 pricing by up to 99%, effective today. Most coverage frames this as a price war. The more interesting part is the engineering that makes these numbers sustainable.
DeepSeek's V4 paper describes a *hybrid attention architecture* that attacks the core bottleneck of long-context inference: the KV cache. Traditional transformers store key-value pairs for every token in the context. At 1 million tokens, this cache alone can fill an entire GPU's memory. V4 introduces two interleaved attention types.
Compressed Sparse Attention (CSA) compresses every 4 tokens into a single KV entry, then selects only the top-k most relevant compressed blocks per query. Heavily Compressed Attention (HCA) goes further, compressing 128 tokens into one entry and running dense attention over the result. The compressed sequence is short enough that dense attention stays cheap.
V4-Pro's KV cache at 1M tokens is 10% (!!) of V3.2's. Single-token inference FLOPs drop to 27% (!!). The model has 1.6 trillion total parameters but only activates 49 billion per token through Mixture-of-Experts routing, the knowledge capacity of a massive model at the compute cost of one thirty times smaller.
MiMo's approach is different but lands in the same place. Xiaomi's team implemented Sliding Window Attention via SGLang HiCache, reducing KV cache data transfer across GPU memory, CPU memory, and SSD to roughly 1/7 (!!) of previous volume. Cacheable tokens expanded by 5x (!!). Combined with expert parallelism optimization and input length bucketing, per-token serving cost dropped enough to make permanent pricing at these levels viable.
V4-Pro now sits at $0.87 per million output tokens. MiMo V2.5-Pro at roughly $3/M output, with Flash variants far below that. A year ago, sub-dollar output pricing meant you were using a small distilled model with real capability tradeoffs. These are frontier-class reasoners with million-token context windows.
Both companies can commit to permanent cuts because the reductions come from the architecture itself. When your attention mechanism physically processes fewer FLOPs per token and your cache occupies a fraction of the memory, the cost to serve is structurally lower. The price follows the cost curve.
🚨 Pavlou Fire Fighting Co. showcased our game-changing wildfire protection at InterAigis 2026 last month.
AXIS cameras pair with LookOut #AI to detect early-stage #wildfires and empower quick response and firefighting in # Greece 🇬🇷 🔥
https://t.co/nxqxi5DKWZ
Thanks to @FortuneMagazine for featuring me this week!
If anyone can build their own software, an interface on top of your data (traditional SaaS) isn't a moat anymore. Meta-apps like Warp, Claude Code, and Codex are where the value is shifting:
https://t.co/dwFft245W0 🔖
AI images are getting harder to spot, but physics still gives them away if you know where to look
Measuring perspective lines can identify AI photos, researchers say
https://t.co/MmpM6oYMT0
VIDEO: A worn-out Pikachu plushie, tired teddy bear or stained stuffed animal can all get a new lease of life at a Japanese laundry service, making beloved toys squeaky clean again. Business is booming at Cleaning Yonmarusan, a regional chain in Yamanashi, west of Tokyo, with customers coming from all over the world for the service.
“You cannot shortcut physics. You cannot deploy half-finished systems into the real world and hope they work.”
Venture capital is moving beyond code because the next tech boom will be built, not programmed
https://t.co/PvpY6LLIIg via @thenextweb
Negative power prices reach all-time high on the Iberian Peninsula in Q1: Spain and Portugal recorded a surge in negative electricity prices in Q1, driven by strong solar generation and relatively low demand… https://t.co/8NayDGGFqn #Photovoltaics#EnergyStorage#RenewableEnergy
🚀 Thrilled to announce: @roboticscats is a Top 10 Finalist in Energy Tech Challengers 2026 – Climate X Wild Card! 🏆
I'll pitch live at @EnergTechSummit in Bilbao, Spain (April 15-16)!
Apple has named John Ternus as its new CEO, succeeding Tim Cook.
Earlier this month, Cook gave WSJ's Ben Cohen an exclusive peek inside the company's archives, showing off early prototypes of the iPod, as part of Apple's 50th anniversary.
Watch more: https://t.co/IxZkwQZhHy
@MuseoGuggenheim As an #EnergyTechChallengers Climate X Wildcard Finalist, @jiansuo presenting FireBird Guard in Bilbao gave us a valuable opportunity to test and refine the concept. 💡
Critics say wind turbines endanger birds but two new studies have now analysed the risk in more detail. What they have found could change the debate.
➡️ https://t.co/s1mfcGiK6v