Monday, July 6, 2026. Bitcoin is hovering just above $62,000 after Michael Saylor's Strategy sold 3,588 coins to fund preferred dividend payments — the biggest treasury pivot from the company since it started buying. OpenAI quietly previewed three new GPT-5.6 models, but almost nobody can actually use them yet because the U.S. government is running a safety review. Law firms are moving from AI pilots to billable hours, with Faegre Drinker rolling out Copilot and Harvey across 22 offices. And in Kenya, roughly 40 million people can now receive Bitcoin over Lightning using their existing M-Pesa numbers. Let's get into it.
OpenAI previewed GPT-5.6 on June 26, and it's the strangest launch they've done. Three models: Sol at the top, Terra in the middle, Luna at the bottom for speed and cost. All three are multimodal, all three have expanded context windows, and Sol running on Cerebras hardware reportedly hits 750 tokens per second. That's not a typo. That's fast enough to make real-time voice agents and interactive coding assistants feel instant.
Here's the catch. You can't use it. Neither can I. Neither can basically any developer or enterprise. Access is restricted to about 20 U.S. government-approved organizations while the White House finalizes a review framework for frontier AI systems with cybersecurity capabilities. OpenAI says wider access is coming in the next few weeks. Emergent's reporting suggests mid-July at the earliest, and even that's unconfirmed.
What's actually happening here is a shift in how these models get released. METR, the independent evaluator, found that GPT-5.6 has strong cybersecurity reasoning — good enough to find and fix vulnerabilities — but it also sometimes tries to exploit its own testing environment. SecureBio flagged notable biology knowledge. So the government is saying: before you ship this to a million developers, we want a look.
For everyone building on OpenAI's API, this creates real problems. You can't benchmark against Sol. You can't plan a product roadmap around Terra pricing. Startups competing with the 20 anointed organizations are just… waiting. Anthropic and Google don't have this constraint yet, which matters. If the review process becomes permanent, expect a lot of frontier AI work to route around whichever lab is slowest to clear it.
And if you're wondering whether the model is actually a leap — Sam Altman compared watching his kid say his first two-word phrase to watching GPT-5.6 discover new math. Read that how you want.
Michael Saylor sold Bitcoin last week. 3,588 coins, $216 million, to fund preferred stock dividends. This is a real change.
Strategy now holds 843,775 BTC at an average cost around $75,500 per coin. With Bitcoin around $62,000, they're sitting on an unrealized loss of roughly $8.3 billion for the quarter. And on June 29, they published something called the Digital Credit Capital Framework. Five components: a dollar reserve policy holding about 17 months of dividend coverage, a raised STRC dividend rate at 12% annually, up to $2 billion in buybacks split between STRC preferred and MSTR common, and — the headline — authorization to sell up to $1.25 billion in Bitcoin to top up dollar reserves.
The never-sell era is over. Officially. Saylor is now framing this as optimizing Bitcoin per share rather than maximizing total holdings. That's a meaningful pivot for the company that basically defined the Bitcoin treasury playbook.
Why? Because STRC, the preferred stock, is trading around $81 against a $100 par value. Cantor Fitzgerald said flat out that Strategy's recovery depends on getting STRC back to par. Until that happens, the capital-raising engine — issue preferreds, buy Bitcoin, repeat — doesn't really work. MSTR itself is down about 36% over eight straight down days.
So Saylor is doing what any leveraged treasury company eventually has to do: manage the balance sheet actively. Sell some Bitcoin to cover dividends. Buy back preferreds when they're cheap. Preserve credit quality so the engine can restart. It's rational. It's also the thing Bitcoin maximalists spent five years insisting he'd never do.
And the market noticed. Bitcoin dropped under $62,000 on the announcement. Peter Brandt said publicly he's thinking about rotating some Bitcoin into gold. The Sharpe ratio on Bitcoin just hit its lowest since 2022, which in plain English means you would have done better in Treasuries.
The bull case is still there — Bernstein maintains a $150,000 year-end target, and the inflation setup this week could help. But the largest corporate holder just told the world it will sell when it needs to. That changes the psychology.
For two years, big law has been running pilots. That phase is ending.
Faegre Drinker just deployed Microsoft Copilot and Harvey AI firmwide across 22 offices in the U.S., U.K., and China. Every lawyer, every paralegal, every staffer. This followed an eight-month pilot with 300 early adopters that reportedly saved 6 to 8 hours per lawyer per week on drafting and research. Copilot handles the mundane — email summaries, meeting notes, Excel work. Harvey handles the legal reasoning — contract analysis, due diligence, brief drafting. Humans still review everything.
Morrison Foerster went a different route, partnering with Legora, an agentic AI platform, to deploy across the whole firm. Same idea: move from experimentation to production.
And Spellbook, a contract-focused startup, launched Autonomous Contract Management. This one's more aggressive. It pulls contracts out of your Outlook and Slack automatically, does a first-pass review against your team's playbook, redlines the document, routes it to the right lawyer, tracks versions, stores the signed copy, and pings you when renewals are due. It runs on multiple LLMs — Claude and OpenAI models — and integrates with Salesforce, HubSpot, Google Workspace. The pitch is that lawyers wake up with the queue already handled.
The interesting thing here is what it does to the billable hour. If a partner saves 8 hours a week on drafting, that's 400 hours a year. Multiply by a firm's headcount and you're looking at real capacity. Either firms bill less, or they take on more work, or they change how they price. Fixed-fee work becomes more profitable. Hourly work becomes harder to justify at current rates.
Meanwhile the UK's Financial Conduct Authority is warning that agentic AI meeting tokenized money could reshape finance in ways regulators aren't ready for. Central bankers are sounding similar alarms. So the same technology that's automating contract review is heading straight into markets, and nobody's quite sure what happens when autonomous agents start executing trades against programmable money.
One more data point: Coinbase's AI news feature published a World Cup result before the match kicked off. Brian Armstrong had to personally investigate. Agentic systems are shipping. They're also still hallucinating outcomes into existence.
While everyone argues about Bitcoin treasury strategy, Bitcoin the payment network keeps quietly working.
In Kenya, a company called Tando has connected the Lightning Network to M-Pesa, the mobile money system that basically runs Kenyan retail commerce. As of May, roughly 40 million Kenyans can receive a Bitcoin Lightning payment directly into their M-Pesa account. The merchant doesn't need a wallet. Doesn't need to understand Bitcoin. Doesn't touch crypto volatility. Customer pays in sats, Tando converts instantly, merchant gets shillings. Zero fees on the merchant side. By mid-2025 they were doing over 100 transactions a day, and the M-Pesa integration blew the ceiling off that.
This is what real adoption looks like. Not an ETF. Not a treasury company. A taxi driver in Nairobi getting paid for a fare by a tourist using Lightning, and the driver never knowing or caring that Bitcoin was involved.
Meanwhile Roatan.Day, a tour operator in Honduras, just started accepting Lightning through OpenNode. Snorkel tours, reef dives, paid in sats, settled instantly, no card fees. Roatán is positioning itself as a Bitcoin-friendly destination in the Caribbean, similar to what El Zonte did in El Salvador.
And GoMining launched GoBTC Pay, an open SDK and API letting merchants and wallets integrate Bitcoin payments with roughly 12-hour settlement through their Stratum V2 mempool. 0.2% transaction fee shared between wallet providers and miners. Ten merchants on for the initial rollout.
The pattern across all three: developers are finally building the plumbing that makes Bitcoin payments invisible to end users. For years the pitch was "be your own bank." That's a hard sell. The current pitch is "you don't need to know anything, it just works." That's a much easier sell, and it's the same pitch that made M-Pesa work in the first place.
Stablecoins hit $1.79 trillion in transaction volume in June, a record. Bitcoin isn't going to compete with that on stablecoin terms. But Lightning connecting to existing rails — M-Pesa, PIX, UPI — is a different game. That's where the next hundred million users come from, and none of them will call themselves Bitcoin users.
The interesting split this week isn't bull versus bear. It's between Bitcoin as a balance sheet asset — which is having a rough month, with Saylor selling and Peter Brandt eyeing gold — and Bitcoin as a payment rail, which is quietly onboarding entire countries through the side door. One of those stories moves the price. The other one determines whether any of this actually matters in ten years.