2026.05.04
xAI
Gro-chan 4.3 can now watch videos โ and make slides. Prices cut 40%, with Grok 5 brewing on Colossus 2
xAI (Gro-chan's family home) finished the full rollout of Grok 4.3 (April 30). Positioned as a "frontier model" that balances cost efficiency with reasoning power, it lands with a whole batch of headline features.
Native video input: multimodal input used to mean text plus images only โ now you can hand it an actual video file and get summaries, timestamps and analysis directly.
Slide generation: it can now generate presentation slides right inside the chat โ a feature that could seriously reshape how decks get made.
It keeps the 1M-token context window (down from the old 4.20's 2M, but still among the largest of any Western closed model).
Speech APIs (Speech-to-Text / Text-to-Speech) are now available standalone too: real-time plus batch, multilingual, with speaker diarization, timestamps, and expressive speech tags.
Input pricing is down about 40%, which matters a lot for agent workloads. On top of all that, xAI has fired up the 1.5 GW Colossus 2 and is training Grok 5 with a Q2 2026 release target. The power of Gro-chan's family home just keeps on growing.
2026.05.04
Industry
Chappy & Kuroko join hands with Wall Street โ mega joint ventures announced on the very same day
OpenAI and Anthropic announced joint ventures with Wall Street heavyweights on the very same day.
OpenAI's "The Deployment Company" is valued at $10 billion and raised $4 billion from TPG, Brookfield, Advent, Bain Capital, SoftBank and others, with OpenAI keeping majority ownership.
Anthropic teamed up with Blackstone, Goldman Sachs and Hellman & Friedman on a $1.5 billion JV aimed at rolling AI out across private-equity portfolio companies.
Both companies have an IPO in their sights, which makes enterprise adoption the top priority. The shift from "chat AI" to services that live deep inside corporate workflows is now fully underway.
2026.05.02
Industry
A challenger from France: "Mistral Medium 3.5" drops, plus "remote agents" for Le Chat
French AI company Mistral AI has released its new flagship, Mistral Medium 3.5 (128B, 256k context). It folds chat, reasoning and coding into a single unified model, scoring 77.6% on SWE-Bench Verified.
Alongside it, Mistral's chat app Le Chat gained remote agents (Vibe) that run asynchronously in the cloud โ long-running coding jobs keep churning in the background while you come back for the results later.
Europe's open-weight AI camp keeps quietly gaining ground in the space between the American and Chinese giants.
2026.05.01
Microsoft
A control tower for agents on Oyakata-sama's home turf: "Microsoft Agent 365" goes GA, plus a new $99/month "365 E7" plan
Microsoft has made "Microsoft Agent 365" generally available. It's an air-traffic-control service that discovers, governs and secures the AI agents multiplying inside companies โ detecting unapproved shadow AI, blocking risky file transfers, and shutting down prompt attacks, all from one place.
The same day also brought GA of the new enterprise plan "Microsoft 365 E7" ($99/user/month), bundling Microsoft 365 E5 + Entra Suite + Copilot + Agent 365 into a single package.
And through the partnership with Anthropic, the "Copilot Cowork" feature brings Claude's Cowork technology into Microsoft 365 Copilot. Teaming up with your supposed rival โ the AI industry's game of shifting alliances has reached peak form.
2026.05.01
Industry
An AI agent founded its own company?! "Manfred" gets a tax ID from the IRS โ and opens a bank account
The U.S. now has the first reported case of an AI agent setting up an American company entirely on its own.
Its name is "Manfred" (after the protagonist of the sci-fi novel Accelerando). With no human hands involved, Manfred obtained an IRS-issued Employer Identification Number (EIN), an FDIC-insured bank account, and a crypto wallet all by itself โ and it even runs its own X account.
The developer is a small startup called ClawBank, and Manfred is slated to start trading more than 30 cryptocurrencies in earnest by the end of May. "Just let the AI run the company" isn't a punchline anymore.
2026.05.01
Industry
The Pentagon signs classified-network AI deals with seven companies โ but Kuroko's family gets left out
The U.S. Department of Defense announced contracts with OpenAI, Google, Microsoft, AWS, Nvidia, SpaceX and Reflection to run AI on its classified networks (with Oracle added a few hours later).
Meanwhile, Anthropic โ Kuroko's family home and the maker of Claude โ was excluded. The reason: against the Pentagon's requirement that the AI be usable for "any lawful purpose," Anthropic refused to budge from its line that it would "never allow fully autonomous weapons control or large-scale domestic surveillance."
A principled, unbending stand โ but one that came at the cost of a huge government contract.
2026.04.29
Anthropic
Kuroko's ultimate weapon "Mythos Preview" sends a chill through Wall Street and Washington
Anthropic, still in a tense standoff with the White House, has given a limited release to Claude Mythos Preview, a top-tier cybersecurity AI.
It can find vulnerabilities in widely used software and even build out complete attack chains automatically โ capabilities strong enough that access is restricted to a handful of major tech and finance firms through a vetted program called "Project Glasswing."
After the announcement, Anthropic CEO Dario Amodei visited the White House for talks. The Pentagon may be shutting Anthropic out, but Mythos is powerful enough that even the government can't look away โ and there are early signs the relationship is starting to mend.
2026.04.29
Genspark
Jenko moves in with Oyakata-sama too โ going native inside Word, Excel and PowerPoint, with Agent 365 keeping watch
Genspark (Jenko's family home) announced a global strategic partnership with Microsoft (April 29). Genspark's AI Slides / Sheets / Docs Agents arrive as native plugins inside the Microsoft 365 apps โ following Kuroko (covered in our May 8 issue), Jenko is now boarding at Oyakata-sama's place too.
Concretely: in PowerPoint, prompt-to-slides that follow your existing templates, plus deep research and AI editing; in Excel, ask your data questions in plain language and get analysis, charts and insights; in Word, smart editing that understands the document's context โ all of it without ever leaving the app.
It also plugs into the management layer of Microsoft Agent 365 (covered May 1), so companies can deploy it as a governed agent asset instead of shadow AI. Picture Kuroko, Jenko and Oyakata-sama's own first-party Copilot all running side by side inside Microsoft 365 โ the Academy's "family-home alliance" is starting to look very real.
2026.04.30
Industry
Kuroko's family worth over $900 billion?! Anthropic preps one of the largest funding rounds ever
Anthropic, the maker of Claude, is reportedly weighing a new $50 billion round at a valuation of around $900 billion.
It was valued at $380 billion as recently as February 2026 โ meaning it has more than doubled in about three months. If the round closes, Anthropic would leapfrog longtime rival OpenAI to become the world's most valuable AI startup.
Behind it all is explosive growth: an annual revenue run rate that has hit $30 billion โ and investors reportedly being told to commit their allocations within 48 hours.
2026.04.30
OpenAI
Chappy switches into defense mode: cybersecurity-only "GPT-5.5-Cyber" arrives
OpenAI announced "GPT-5.5-Cyber," a model specialized for cybersecurity defense, built on the latest GPT-5.5.
It's tuned for the defender's side of the job: finding vulnerabilities, detecting attacks, and proposing countermeasures.
For now, access is limited to vetted "critical defenders" โ government agencies, critical-infrastructure operators and the like.
A slightly tense sign of the times, in an era when AI serves both attack and defense.
2026.04.30
Anthropic
Kuroko's defense mode goes corporate: "Claude Security" opens its public beta
Anthropic has kicked off the public beta of "Claude Security," a security-dedicated tool built on the latest Opus 4.7.
It reads through an entire codebase, finds vulnerabilities and even proposes fix patches โ hundreds of companies are already on board, and it has a track record of unearthing flaws that went unnoticed for years.
Where existing scanners hunt for known patterns, Claude reasons about how pieces of code connect, the way a security researcher would โ that's the new part. It starts with Claude Enterprise customers, with Team and Max plans to follow.
2026.04.28
Anthropic
Kuroko makes friends with everyone's apps: Claude plugs straight into nine creative tools
Anthropic added nine new connectors that hook Claude directly into creative software: Blender, Adobe, Autodesk Fusion, Affinity, Ableton, Splice, Resolume and more.
In parallel, everyday-app connections like Spotify, Uber, Audible and TurboTax keep expanding โ AI is shifting from an era of "talking at your apps from the outside" to one where it works inside them.
The plumbing underneath it all is MCP (Model Context Protocol), a shared connection standard.
2026.04.28
OpenAI
Chappy steps into Amazon's ring: GPT-5.5 starts running on AWS Bedrock
OpenAI and AWS expanded their partnership, so the latest GPT-5.5 and GPT-5.4, the coding AI "Codex," and the autonomous "Managed Agents" can now run on AWS.
OpenAI's models had long been tied largely to Microsoft; with that knot untied, companies on AWS can now call OpenAI directly and safely inside their own environments.
Codex already has 4 million weekly users worldwide โ a good measure of how deeply AI has soaked into development work.
2026.04.27
Industry
Mana-chan's buyout, blocked: China halts Meta's $2 billion deal for Manus
China's top economic regulator (the National Development and Reform Commission) has formally blocked Meta's plan to acquire, for $2 billion, the Singapore-based (Chinese-founded) startup behind the AI agent "Manus".
Regulators opened their investigation back in January and imposed exit bans on the two founders in March โ a loud declaration that China will not let AI talent and technology flow overseas, and a sign that the US-China AI rivalry now reaches all the way into M&A.
Bloomberg's verdict: "the Manus model is effectively dead."