WEBVTT NOTE The Rundown — nextbig.dev daily audio edition, 2026-06-26 1 00:00:04.500 --> 00:00:09.220 OpenAI shipped its strongest model today, and almost no one can use it. 2 00:00:09.220 --> 00:00:22.660 It's Friday, June twenty-sixth. Here's the rundown: a frontier model that launched to the government first, the three-tier family underneath it, and what it means if you build on someone else's API. 3 00:00:22.840 --> 00:00:38.120 OpenAI previewed GPT-5.6 this morning. Three models: Sol, Terra, and Luna. The flagship, Sol, went to about twenty trusted partners through the API and Codex, and to nobody else. 4 00:00:38.120 --> 00:00:52.440 And that's where the story is. OpenAI says it showed the models to the US government first, and at the government's request it opened with that small list. Bloomberg reported it came under pressure from the Trump administration. 5 00:00:52.440 --> 00:01:12.440 Sol is genuinely strong. Its new ultra mode, where it splits a task across subagents, lands around ninety-two percent on a coding-agent test. GPT-5.5 was eighty-three. Same price, five and thirty dollars a million tokens. 6 00:01:12.440 --> 00:01:23.560 It's good. But OpenAI's own safety review rates it High for cyber and bio risk, one step under critical, and that rating is the reason it's gated. 7 00:01:23.560 --> 00:01:27.960 So a frontier model is now released like a controlled technology. 8 00:01:27.960 --> 00:01:40.200 Previewed to the state, then handed to a list the state signed off on. If you build on a frontier API, the model you depend on can now be pulled by a government order, on top of the usual rate limits. 9 00:01:40.200 --> 00:01:42.360 What do you do about it. 10 00:01:42.360 --> 00:01:51.560 Keep a second model wired in that you can actually call today, and treat 'broad availability in a few weeks' as a promise with no schedule attached. 11 00:01:53.900 --> 00:02:02.060 This has been building all week. Wednesday, Claude rationed capacity at peak hours, it just couldn't serve the demand. 12 00:02:02.060 --> 00:02:16.900 Today OpenAI rations the other way, by permission. Two labs, two kinds of scarcity. And Tuesday we said the coworker fight had moved to distribution when Claude landed in Slack. Today it moved again, to access. 13 00:02:16.900 --> 00:02:19.180 The export-control thread, too. 14 00:02:19.180 --> 00:02:27.260 The same logic that gates chips going abroad is now gating a model's release at home. We tied it all together in today's edition. 15 00:02:27.440 --> 00:02:41.040 To the tape. We're long Microsoft, because OpenAI's gated model still reaches enterprises through Azure and Copilot first. We're watching Alphabet and Nvidia. 16 00:02:41.040 --> 00:02:55.920 Highest conviction is Microsoft. The falsifier: if companies adopt GPT-5.6 direct and bypass Azure, or the limited rollout stalls the money into next year, we're wrong. Horizon, twelve months. 17 00:02:55.920 --> 00:03:01.280 The tape is the desk's scorecard, not advice. 18 00:03:04.520 --> 00:03:06.760 Quick break — two from the desk. 19 00:03:06.760 --> 00:03:20.760 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. 20 00:03:20.760 --> 00:03:33.800 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. 21 00:03:38.310 --> 00:03:52.790 Our call: within six months, this becomes the default. Another top lab, Anthropic or Google, ships a flagship the same way. Government preview, limited list, then everyone else. 22 00:03:52.790 --> 00:04:00.550 What proves us wrong: if by the end of December no other top lab gates a launch like this. It settles December twenty-sixth. 23 00:04:00.550 --> 00:04:07.990 Today the one thing standing between you and the best model was clearance. That's the rundown.