Here's what's moving today. Google dropped Gemini 3.7 Flash, its new workhorse coding model, at aggressive introductory pricing. The SEC surprised everyone with a sweeping crypto rule proposal after quietly cancelling the meeting that was supposed to vote on it. Bitcoin is stuck in a six-week range around $64,000 while global bond yields hit multi-decade highs. Spot Bitcoin ETFs pulled in another $189 million Tuesday, pushing August inflows close to $1 billion. And Metaplanet is planting a flag in the US with a $135 million bitcoin treasury deal on Nasdaq. Let's get into it.
Google shipped Gemini 3.7 Flash this week, and the headline number is the pricing: 75 cents per million input tokens, $3.75 per million output tokens through year-end. After December 31st it doubles to $1.50 and $7.50. That's Google buying market share, plain and simple.
The benchmarks are actually interesting though. On FrontierCode 1.1 it jumped from 34.4% to 43.6%. On DeepSWE, from 49% to 65.3%. On document processing, from 22% to 34%. These aren't small deltas for a point release. Google is clearly pushing Flash to eat into work that used to require Pro-tier models.
And notice what's missing: Gemini 3.5 Pro. Still delayed. Bloomberg pointed out Google is shipping a better Flash while its flagship stays in the oven. That's a tell. Either Pro isn't hitting internal targets or they're rethinking the whole tier structure. Meanwhile OpenAI, according to CoinDesk, is trailing Anthropic on revenue, losses are widening, and Altman reportedly paused frontier reinforcement-learning training to tighten safety controls. The frontier race has genuinely gotten weird. The lab everyone assumed would run away with it is pausing. Google is pushing mid-tier. Anthropic is quietly winning on developer mindshare.
3.7 Flash is now the default in the Gemini app, powers the Spark agent inside Google Workspace, and is a selectable model in AI Mode for Search for Pro and Ultra subscribers. Google is threading this model through every surface they own. If you're building agents or doing knowledge work in finance, law, or biosciences, this one is worth benchmarking against Claude and GPT this week. The price gap is real.
The coding agent space had a wild 48 hours. Cursor, which is now owned by SpaceX, launched Origin, a direct competitor to GitHub. They're capitalizing on recent GitHub outages and pitching interoperability, so you can sync repos between the two. GitHub has 180 million developers, so this is a long war, but the fact that Cursor thinks it can even show up to that fight tells you how much the coding tools market has reshuffled.
Warp launched Warp Factories, an out-of-the-box AI software factory that plugs into Codex, Claude Code, Linear, Jira, Slack. It tracks token spend, monitors agent performance, and targets mid-size companies that can't build this stuff themselves. UiPath released Maestro Flow, which is basically enterprise orchestration for coding agents, so your Claude Code prototype and your production deployment are the same artifact.
And Light Cloud released a Model Context Protocol server that lets Claude Code deploy full-stack apps without leaving the chat window. You describe what you want, Claude provisions the app, wires the database, ships the code, and hands back a live URL.
The most honest data point though comes from Anthropic itself. They ran Claude Code unsupervised on their own internal apps. It opened 388 pull requests. Humans merged 180. That's a 46% merge rate, which sounds impressive until you remember reviewers had to read all 388. The real question isn't merge rate, it's whether the time saved on the accepted PRs outweighs the time spent rejecting the other 54%. Nobody has published that number yet.
And on the security side, Bybit told CoinDesk that AI helped it save $700 million a year after losing $1.46 billion to North Korean hackers. First large exchange to actually put a number on the AI-for-security claim that bitcoin developers have been making all month. Take it with salt, but the direction is clear: agents are moving from demo to production, and the metrics are starting to get real.
The SEC just did something genuinely unexpected. On Monday they cancelled a scheduled meeting. On Tuesday they published Regulation Crypto anyway, their first major crypto rulemaking proposal. It arrived days after the Senate failed to pass the CLARITY Act.
The proposal creates a safe harbor so certain tokens aren't automatically treated as investment contracts, and it carves out exemptions for token issuance. Hester Peirce, unsurprisingly, called it an important step forward from what she called inapt legacy rules.
But here's where it gets contentious. CryptoSlate pointed out the SEC's proposal deliberately treats tokens as freely tradeable the moment a buyer acquires them, unless the issuer says otherwise. The Senate draft floating around would block exactly that: it would restrict day-one insider sales. Insiders in early-stage token projects know more than the public, and their incentives don't always line up with retail. So the SEC is opening a door the Senate was trying to close. That gap is going to be the whole fight for the next six months.
Former New York Governor Andrew Cuomo, of all people, is out there arguing CLARITY has to pass, that the US is falling behind. The American Bankers Association wants CLARITY strengthened, not killed. Meanwhile the FASB proposed conditions under which stablecoins could be treated as cash equivalents on corporate balance sheets, with the requirement that holders have direct issuer redemption rights and one-to-one liquid reserves. That's a big deal for treasury departments considering stablecoins.
And in a story that got less attention: China added eight more banks to its digital yuan network this week, tripling the e-CNY operator count this year. While the US argues about safe harbors, China is just wiring up the rails.
Bitcoin is stuck. Price is holding around $64,000, tagged $65,000 briefly, but volatility has dropped to multi-year lows. CoinDesk's framing is right: traders who used to live off bitcoin's chaos are chasing 5x and 10x payoffs elsewhere. VanEck says bitcoin is flashing 8 of 12 capitulation signals, but they don't think the bottom is in yet. Historically these setups produced below-average 90 and 180-day returns.
Meanwhile, long-term bond yields are hitting multi-decade highs as US debt approaches $40 trillion and AI hyperscalers keep issuing paper. That's testing bitcoin's hedge narrative in real time. BlackRock's latest report says bitcoin has largely purged the froth that preceded its 50% drop from $126,000, and that it still holds up as a diversification asset.
On the ETF side, spot bitcoin ETFs pulled $189 million Tuesday. August inflows are approaching $951 million. BlackRock's IBIT led with $160 million, Fidelity's FBTC added $112 million. Those two funds took 91% of the day's bitcoin ETF flows. Institutional concentration is intense.
And the Q2 13F filings tell a nuanced story. Jane Street disclosed over $1 billion in spot bitcoin ETF exposure. UBS boosted its IBIT call options 24-fold to 1.95 million underlying shares while cutting puts by 53%. JPMorgan grew IBIT holdings from about 8.4 million shares to 10.6 million. Banks are buying. Hedge funds trimmed. Sovereigns held.
On the payments side, the ground shifted quietly. Square hit 1 million US merchants accepting Lightning bitcoin payments. A new business turns it on every 8 seconds, according to Square. Fees are waived through 2026, and there's a 5% bitcoin-back rewards program. Amboss launched an affiliate program paying 15 to 20% of platform fees for a year, paid in bitcoin over Lightning, with a flat 0.5% merchant rate that settles in USDC, USDT, or BTC using Taproot Assets. And Metaplanet is bringing 2,100 BTC and $2.5 million cash into a US-listed treasury vehicle through a $135 million nanocap deal on Nasdaq, potentially unlocking up to $3.4 billion if warrants get exercised.
The pattern is consistent. While price sits still, the rails keep getting built out.
One thing to notice: the SEC just proposed the crypto framework Congress couldn't pass. If Regulation Crypto sticks, the US will have gotten its rules from an agency, not a legislature, and every future administration can rewrite them by decree. That's a very different regulatory foundation than what CLARITY would have built. Watch what the Senate does next. That's the real story.