Bitcoin punched back up near $64,000 today, erasing the week's dip, while spot Bitcoin ETFs bled another 95 million dollars on Thursday. Google made AlphaEvolve generally available and rewrote its Managed Agents API. Cognition dropped SWE-1.7, Meta launched Muse Spark 1.1, and Grok 4.5 landed with aggressive pricing. Circle got final approval for a national trust bank charter. And New Hampshire's trailblazing 100 million dollar Bitcoin-backed bond just got killed 3 to 2 by the state's executive council. Four things worth your time today.
Google had a big week for developers, and it's worth breaking down because two announcements actually matter.
First, AlphaEvolve is now generally available on the Gemini Enterprise Agent Platform. This is the code optimization agent DeepMind has been quietly using inside Google to redesign TPU circuit layouts, cut Spanner's write amplification by around 20%, and shrink software storage footprints by about 9%. You give it a seed program and an evaluator script, and it explores the search space, mutating solutions and scoring them until it finds something better than what your engineers wrote. It's been used on quantum circuits, natural disaster prediction, and semiconductor design. Now anyone with a Google Cloud contract can point it at their own algorithms.
The second piece is more practical for people building today. Google expanded Managed Agents in the Gemini API with four features that push agents from demo territory toward actual production. You can now run agents asynchronously in the background — set background to true, get a task ID, poll it, reconnect later. That kills the fragile long-lived HTTP connections everyone's been hacking around. Managed Agents can now connect directly to remote Model Context Protocol servers, so your agent can talk to internal databases and private APIs without a custom proxy. Custom function calling now sits alongside the built-in sandbox tools, and you can rotate credentials mid-session without blowing away sandbox state.
The caveats matter. This is still preview. No versioning, no subagent nesting, and you need real governance around which MCP servers agents can reach. But the direction is clear: Google is trying to make agent infrastructure boring and durable, which is what enterprises actually want. OpenAI and Anthropic have been eating Google's lunch on developer mindshare, and this is a credible response.
One more thing worth mentioning — DeepMind also shipped a Predicting the Past skill for historians working with ancient Greek and Latin inscriptions, wiring Gemini into their Ithaca and Aeneas models. Niche, but a nice reminder that specialized agent skills are becoming the actual product surface.
The coding model market got very crowded this week. Let's go through it fast.
Cognition released SWE-1.7. Built on a Kimi K2.7 base with heavy reinforcement learning post-training, distributed across three continents with fault-tolerant training. The interesting technical bit is self-compaction — the model summarizes its own working state so it can resume from a summary and stretch rollouts to six hours. That's the ceiling that matters for autonomous coding work. They also alternate between unconstrained and budget-constrained training phases, which is a smart way to keep the model concise without killing long-horizon behavior. One honest note from Cognition: as reasoning improves, the model touches more files and writes more tests. That's not a bug, that's what a competent engineer does, but it will cost you tokens.
Meta launched Muse Spark 1.1. Multimodal, multistep reasoning, priced at 1.25 dollars per million input tokens and 4.25 per million output. That's right on top of Claude Haiku 4.5 and GPT-5.6 Luna. Meta is late here and they know it — this is a catch-up product with sharp pricing.
SpaceXAI shipped Grok 4.5, trained alongside Cursor, priced at 2 dollars in and 6 out. They're claiming roughly 2x token efficiency at about 80 tokens per second, and it's already live in Cursor on all plans.
And OpenAI's GPT-5.6 family — Sol, Terra, and Luna — is now in GitHub Copilot. Sol for heavy reasoning on large codebases, Terra as the everyday default, Luna for cheap fast work. Enterprise admins have to explicitly turn on the policy, which is off by default.
IBM meanwhile is going a different direction with Bob — multi-agent orchestration with cost analytics they're calling Bobalytics, and premium packages for mainframe, IBM i, and Java 25 migration. Not exciting, but that's where the enterprise money actually is.
The read here: frontier coding intelligence is commoditizing fast, prices are compressing, and the differentiation is moving to long-horizon reliability, tool use, and integration. If you're paying flat rates for a single vendor right now, you're probably overpaying by summer's end.
Bitcoin climbed back to around 64,000 today, retesting the level it failed to break on Monday. If it clears there, the June 15 peak at 67,250 is the next obvious target. A long-term MACD gauge just flipped bullish, which the technical crowd is reading as fuel for more upside.
But the derivatives picture is more interesting than the spot price. Bitcoin has now spent 307 days in the 60,000 to 70,000 range — the third longest consolidation in any 10,000 dollar band in Bitcoin's history. That's a compressed spring, and traders are positioning for the release. CME options open interest remains skewed heavily toward puts, which tells you institutions are still buying downside protection rather than chasing calls. Today also saw about 23,000 options expire with roughly 1.5 billion dollars in notional and a put-call ratio right around 0.97 — basically balanced, with max pain sitting near 62,000.
Meanwhile spot Bitcoin ETFs lost another 95 million dollars on Thursday, and ether funds snapped a five-day inflow streak with 52 million out. That's the more worrying signal. Institutional flows have been the story of this cycle, and when they turn, price tends to follow eventually. CoinDesk flagged that redemption requests in the 2 trillion dollar private credit market hit 15.6 billion in Q2 — a much bigger risk-off signal than the ETF outflows, and worth watching as a macro tell.
There's also a currency split happening that most people are missing. Bitcoin is strong in dollar terms but lagging in yen because the yen is ripping on Japanese intervention fears. If you're a Japanese buyer, Bitcoin doesn't feel like it's ripping at all. That matters because Japan just became the story on the demand side — more on that next.
The honest read: this is a market waiting for a catalyst. Leverage keeps building, the range keeps compressing, and the longer it grinds, the sharper the eventual move.
The most interesting Bitcoin story of the day isn't the price — it's Japan quietly becoming the most bullish jurisdiction on earth.
Japan's government is rolling out an "invest locally" plan that CoinDesk says is likely to spur demand for assets like Bitcoin and gold. Combined with yen intervention pressure and negative real rates, you're setting up conditions where domestic capital needs somewhere to go that isn't yen.
Enter Metaplanet. On July 13, they close their acquisition of Siiibo Securities for about 2.1 billion yen, roughly 13 million dollars. They're renaming it Metaplanet Securities and launching what they're calling Project Nova — a Bitcoin-focused brokerage inside Japan's regulated financial system. The acquisition buys them a Type I Financial Instruments Business Operator license, which is the golden ticket to publicly distribute Bitcoin-linked yield products, BTC-backed bonds, and tokenized securities to Japanese retail. Initial target: their existing 250,000 shareholders.
At the same time, Metaplanet is working with JPYC — the yen-pegged stablecoin — and Progmat on Bitcoin-backed digital credit products. Think 24/7 settlement, daily interest accrual, blockchain-native corporate bonds using BTC as collateral. Nothing has launched yet, but the direction is unmistakable: they're building a full Bitcoin financial services stack for Japan.
Metaplanet isn't the only one. A Japanese lender called CRYL just launched Bitcoin-backed loans up to about 6.2 million dollars for individuals and businesses. That's the credit-market plumbing that turns Bitcoin from a speculative asset into productive collateral.
Here's the honest picture though. Metaplanet's business model is under real strain. Their Bitcoin options revenue collapsed 41% quarter over quarter to about 11 million dollars because low volatility around 60,000 has crushed premiums. Their market-to-NAV premium disappeared, so equity issuance is unprofitable and they've pivoted to debt. They hold 43,000 BTC worth about 2.6 billion dollars, third largest corporate holder behind Strategy and Twenty One Capital, but they've set a 100,000 BTC target by year-end. That requires roughly 9,500 BTC a month in the back half. Last quarter they added 2,823. The math is aggressive.
So the setup is genuinely interesting. A jurisdiction pushing domestic investment. A regulated Bitcoin brokerage about to launch. A stablecoin and tokenization stack being built. And a lead company that has to execute at a pace the market currently doesn't believe in. If Metaplanet pulls it off, Japan becomes the model for how corporate Bitcoin treasuries expand into full financial services. If they don't, it becomes a cautionary tale about financialization outrunning fundamentals.
One prediction: the next twelve months of Bitcoin adoption get written in Tokyo, not Washington. While the CLARITY Act crawls through committee and the SEC and CFTC sit understaffed, Japan is quietly assembling regulated brokerages, backed loans, and stablecoin rails around Bitcoin. Watch what ships, not what gets debated.