Big day across the board. OpenAI shipped GPT-5.6 and it's now the default model powering Microsoft 365 Copilot across Word, Excel, PowerPoint and Chat. They also launched ChatGPT Work, an agentic platform aimed squarely at the enterprise. Bitcoin is wobbling around $62,600 as Trump reinstated the Hormuz blockade and traders piled into bets on a July Fed rate hike. June CPI came in soft, down 0.4%, which may cool those hike bets by afternoon. Strategy went a third straight week without buying any Bitcoin, instead padding its cash reserve to $3 billion. And a solo miner running $150 of hardware just hit a $200,000 block. Let's get into it.
OpenAI's GPT-5.6 launch is the biggest AI story of the week, and it's less about raw benchmarks than about distribution. Microsoft is making it the default across all of Microsoft 365 Copilot. That means every Word draft, every Excel analysis, every PowerPoint deck touched by Copilot is now routed through 5.6. The lineup has three tiers: Sol is the flagship at $5 per million input tokens and $30 output, Terra is the mainstream option at $2.50 and $15, and Luna is the cheap high-volume model at $1 and $6. Alongside the model release, OpenAI shipped ChatGPT Work, an autonomous agent platform that connects to Microsoft 365, Google Drive, Slack and Notion, and can run multi-step workflows for hours. Sol includes an ultra mode that coordinates four sub-agents in parallel. Now here's the interesting bit for anyone tracking the security angle: Sol scores 73.5% on ExploitBench versus 47.9% for GPT-5.5. OpenAI says they ran roughly 700,000 GPU hours of automated red-teaming before shipping this. That's a real number. The enterprise question is trust. Would you let an autonomous agent operate for hours inside your sensitive systems? OpenAI is betting that governance tooling, real-time monitoring, and stress testing are what closes deals now, not benchmark bragging rights. Anthropic is the obvious competitor here, and the fight for enterprise contracts is going to be won on reliability, not IQ.
Speaking of agentic AI in regulated environments, Norm AI just raised a $120 million Series C at a $1.2 billion valuation, led by Khosla Ventures. The investor list is a who's-who of traditional finance: Blackstone, Bain Capital, Coatue, Vanguard, New York Life, TIAA. Total funding is now over $260 million in under three years. What Norm actually does is interesting. It embeds legal rules into AI agents, and its affiliated law firm, Norm Law, is AI-native. It bills clients on outcomes, not billable hours. Its clients collectively manage over $30 trillion in assets. The agents draft documents, do compliance reviews, and can even supervise other AI agents operating in regulated workflows. That supervisory layer is the real product. Meanwhile in Helsinki, a new firm called Brahe launched claiming to be the first Nordic AI-first law firm. Their pitch: work that used to take three weeks now takes three days, with the AI as the operating system and lawyers as the final judgment layer. Two very different strategies, same underlying bet. The billable hour is a dead man walking in regulated professional services. If a task is repetitive and rule-bound, an agent can do it, and someone will price it based on the outcome, not the clock. What's notable is that the money flowing into Norm isn't from crypto-adjacent VCs chasing narratives. It's pension funds and insurance giants. They're not investing in AI as a story, they're investing because they're going to be Norm's customers.
Strategy, formerly MicroStrategy, has now gone three consecutive weeks without buying any Bitcoin. Instead, Michael Saylor's company added $450 million to its cash reserve last week, bringing the USD buffer to $3 billion. That's a real shift. In the two weeks prior, Strategy actually sold 3,588 BTC, roughly $216 million worth, to help fund preferred dividend obligations. The company still holds 843,775 BTC, acquired for an aggregate $63.69 billion. And the ATM activity is telling: between July 6 and 12, Strategy sold 4.8 million MSTR shares for $466.7 million in net proceeds. None of that went into Bitcoin. All of it went to the cash cushion. Why? The $3 billion reserve now covers roughly 20.4 months of preferred stock dividends and debt interest, against about $1.76 billion in annual obligations. Strategy's enterprise multiple to net asset value sits around 1.02, meaning the stock is trading essentially at book. When your mNAV premium collapses, you can't just issue more equity to buy more BTC, that game only works when the market pays you a premium. So Saylor is defending the balance sheet. This matters for the whole Bitcoin treasury company thesis. The playbook of infinite share issuance to buy Bitcoin only works in a bull market with a fat mNAV premium. Strategy is now showing us what phase two looks like: preserve liquidity, protect dividends, wait. If Bitcoin keeps sliding, the $3 billion reserve might not be enough, and we could see the option to sell up to $1.25 billion more BTC get exercised. The never-sell mantra is officially retired.
On the Bitcoin infrastructure side, Lightning had a genuinely good week. Polymarket added Lightning deposits via the Spark protocol, which kills the 10 to 60 minute on-chain confirmation window. Deposits credit in under a second, and custody stays with the user until a trade executes. Polymarket is now channeling Bitcoin liquidity into prediction markets right as that whole category crossed $50 billion in monthly volume during the World Cup. That's real trading volume, larger than a lot of traditional sportsbooks. BTC Inc. rolled out Lightning across its entire operation using BTCPay Server, covering ticketing, point of sale, and e-commerce ahead of Bitcoin 2026 in Las Vegas. And a new app called Radar Chat launched on iOS and Android, combining Signal-protocol encrypted messaging with self-custodial Lightning payments inside chats. You type a sats amount, hit send, done. The team has ties to Cake Wallet and its 2 million users. Meanwhile GoMining launched GoBTC Pay, which offers instant checkout but routes settlement through its own mining pool with a 12-hour on-chain settlement window. That one's worth watching skeptically. It solves UX but reintroduces a centralized settlement point, which is exactly the tradeoff Lightning was built to avoid. Compare that with the open BTCPay Server plus Lightning stack, and you can see two very different philosophies competing for the merchant integration layer. My read: Lightning integration is finally reaching the phase where builders stop asking whether users will use it and start asking which apps they're already inside. Prediction markets, messaging, event ticketing, those are the wedges.
One thing to watch: Strategy is signaling that the era of aggressive Bitcoin accumulation via equity issuance has a ceiling, and that ceiling is your mNAV premium. When it collapses, the model changes. Every corporate Bitcoin treasury out there just got a preview of what discipline looks like.