WEBVTT NOTE The Rundown — nextbig.dev daily audio edition, 2026-07-09 1 00:00:04.500 --> 00:00:15.540 It's Thursday, July ninth. Two frontier models reached the public today, on the same afternoon. And the more interesting fact is how each one got there. 2 00:00:15.540 --> 00:00:28.260 Here's the rundown. Why OpenAI's strongest model spent twelve days behind a government gate, how Grok matched its tier at a quarter of the price, and why the model itself stopped being the scarce thing. 3 00:00:28.440 --> 00:00:47.520 OpenAI switched GPT-5.6 on for everyone this morning, in three trims it calls Sol, Terra, and Luna. This is the same model it launched on June twenty-sixth to about twenty government-approved partners and no one else. 4 00:00:47.520 --> 00:01:05.400 That's the part nobody voted for. It spent twelve days locked to vetted organizations after the White House's cyber and science offices asked OpenAI to hold the launch, citing the cyber capabilities of the top model, Sol. Voluntary on paper. In practice, it worked like preclearance. 5 00:01:05.400 --> 00:01:27.720 And by the afternoon, xAI's Grok 4.5 was public too. Elon Musk calls it an Opus-class model, faster and cheaper, two dollars a million in and six out, against the five and twenty-five Anthropic charges for Opus 4.7. 6 00:01:27.720 --> 00:01:41.960 The commoditization is printed right on the price sheet. Luna, the cheapest trim, is a dollar a million tokens in. Two labs shipped comparable capability on one afternoon. That's what it looks like before anyone calls it that. 7 00:01:42.140 --> 00:01:55.900 This is where the whole week converges. The value slid off the model into the memory and the agent around it, and today the model itself finished the trip into a tiered product line. 8 00:01:55.900 --> 00:02:17.100 TechCrunch put the mirror image on it this morning. Nvidia is a victim of the marketplace it created. Compute is deflating, its stock down about fifteen percent since May, while the memory beside the chip is up roughly tenfold in a year and Micron has nearly tripled. Everyone now builds their own silicon, and no one builds their own memory. 9 00:02:17.100 --> 00:02:33.100 So if the model is the commodity, the moat is who you're allowed to sell it to, and who you can reach. OpenAI answered the second half the same morning, shipping ChatGPT Work, its bid for the enterprise seat. 10 00:02:36.080 --> 00:02:51.280 To the tape. We're holding the Micron long. Today's compute story reinforces it from the other side: everyone builds silicon, no one builds DRAM, and the memory line is the one that keeps climbing. 11 00:02:51.280 --> 00:03:07.040 We're sharpening the Nvidia watch, but not moving to a short. The demand is real, two frontier launches are more inference, not less. The pressure is on pricing power, as every hyperscaler ships a good-enough custom chip. 12 00:03:07.040 --> 00:03:21.520 And new to the book: Meta, on watch. It begins producing its own AI chips in September. That custom-silicon wave is the mechanism doing to the GPU what today did to the model. 13 00:03:21.520 --> 00:03:26.720 The tape is the desk's scorecard, not advice. 14 00:03:31.160 --> 00:03:33.240 Quick break — two from the desk. 15 00:03:33.240 --> 00:03:47.240 One we know well: vote dot direct. If you're on an H O A or a board, it runs your elections digitally — secure, verifiable, no paper, no clipboard in the lobby. Point your council to vote dot direct. 16 00:03:47.240 --> 00:03:59.560 And if this is your ten minutes of A I for the day, get the written edition too. The full wire, free, every morning — leave your email at nextbig dot dev. 17 00:04:04.310 --> 00:04:28.790 Our call: within the next six months, at least one more frontier model, from OpenAI, Anthropic, Google, xAI, or Meta, gets gated to government-approved users or held for a review before its public release. The June gate was the first instance of a preclearance regime, not a one-off. 18 00:04:28.790 --> 00:04:38.910 What proves us wrong: if by January no other frontier model is gated that way, and the GPT-5.6 gate stays the only instance on the record. 19 00:04:38.910 --> 00:04:58.630 The model got cheaper, got cleared, and got a price war on the same afternoon. Luna costs a dollar a million tokens, and it shipped this morning. A gigawatt to run it is still four years deep in a queue. That gap is the market. That's the rundown.