WEBVTT NOTE The Rundown — nextbig.dev daily audio edition, 2026-06-24 1 00:00:04.500 --> 00:00:12.380 Claude went dark for two hours and thirty-seven minutes yesterday, and that's the third week running it's failed at the exact hour US teams ship. 2 00:00:12.380 --> 00:00:26.180 It's Wednesday, June 24, 2026. Here's the rundown. We've got a provider rationing capacity, China taking the Top500 on CPUs alone, and a 744-billion model running on a Mac. 3 00:00:26.360 --> 00:00:44.520 Anthropic logged elevated error rates starting just after two in the afternoon UTC on Monday. By the close of the day Claude had been degraded two hours and thirty-seven minutes, across the website, the API, the Console, Claude Code, all of it at once. 4 00:00:44.520 --> 00:01:00.040 And one bad day is noise. I'd let one bad day go. But June 16th every Sonnet and Opus ran near a ten percent error rate, and on the 13th they suspended two models outright. No root-cause writeup for any of it. 5 00:01:00.040 --> 00:01:10.280 The errors are 529 overloads, clustered at peak US hours. Anthropic told Fortune demand has outrun what its infrastructure can serve. 6 00:01:10.280 --> 00:01:23.560 Right, so that's the tell. A 529 isn't a crash, it's capacity rationing. They're turning people away at the door because the room is full, and the fix is more capacity through Amazon and Google that isn't online yet. 7 00:01:23.560 --> 00:01:35.400 Here's the exposure. A SemiAnalysis count in February put Claude Code at roughly four percent of all public GitHub commits. More than a hundred thirty-five thousand a day. 8 00:01:35.400 --> 00:01:48.920 So a real slice of the world's merges now leans on one provider that throttles exactly when you're trying to ship. Put Claude in your merge gate and your release cadence inherits their overload curve. I've been paged for less. 9 00:01:48.920 --> 00:01:51.240 What do you do tonight if you're building. 10 00:01:51.240 --> 00:02:04.200 Stop treating Claude as a hard dependency in any automated path. Put a fallback behind a router so a 529 reroutes instead of blocking the build. The hedge actually landed this week. 11 00:02:04.200 --> 00:02:16.040 Unsloth's day-zero weights for GLM-5.2. A 744-billion open model on a 256-gig Mac, two-bit, MIT licensed. 12 00:02:16.040 --> 00:02:29.320 Single-digit tokens a second, so it's a backstop, not a swap. But a slow local model that answers beats a fast remote one returning 529. No rate limit, no status page to refresh. 13 00:02:30.000 --> 00:02:34.800 And Azure quietly added Fireworks open-weight serving to Foundry the same day. 14 00:02:34.800 --> 00:02:47.920 Which is the cleanest version of this: a fallback one config away inside the cloud you already pay. Every outage teaches another team to wire an escape hatch, and nobody rips those out afterward. 15 00:02:48.300 --> 00:03:01.980 China retook the Top500 at 2.2 exaflops, beating El Capitan's 1.8. First machine to sustain over two exaflops of double precision without a single GPU, on a custom 304-core chip. 16 00:03:01.980 --> 00:03:14.220 And that's the export-control story in one number. The embargoes didn't cap peak compute, they pushed China toward CPU-dense designs that route around the banned parts entirely. 17 00:03:14.220 --> 00:03:22.140 On the power side, Microsoft locked twenty years of gas for datacenters, and Canada lined up ten reactors by 2040. 18 00:03:22.140 --> 00:03:33.500 Note Canada's strategy has no money attached. It points vaguely at an infrastructure bank. That's a posture, not a build order, with nuclear still thirteen percent of their grid. 19 00:03:33.500 --> 00:03:43.100 But the gas commitment is real concrete. A hyperscaler conceding AI load outpaces the grid and buying baseload on a two-decade contract. 20 00:03:43.100 --> 00:04:04.140 That's the shift worth holding onto. The constraint on scaling inference is moving from chips to firm electrons, and they're financing it years ahead of the load. Watch CXMT too, the Chinese memory maker filing to IPO. A fourth serious DRAM and HBM supplier changes the cost floor under every buildout. 21 00:04:04.320 --> 00:04:13.520 The GLM-5.2 build we mentioned needs 1.5 terabytes full, but the two-bit drop fits in 239 gigs at about eighty-two percent accuracy. 22 00:04:13.520 --> 00:04:27.320 Frontier-class open inference on unified memory. The vendor claims parity with Opus and GPT-5.5, and not one of those benchmarks is independent yet. So enjoy it, verify it. 23 00:04:27.320 --> 00:04:37.520 Smaller theme repeating today. A 35-billion model topped a forecasting leaderboard, and a 3-billion model claims it beats Opus 4.5 on reasoning. 24 00:04:37.520 --> 00:04:52.240 Both self-reported, both arxiv-fresh. But the signal is real enough: on bounded tasks, stop reaching for a frontier model. A 35-billion serves cheaper and faster and you'll never feel the gap. 25 00:04:52.240 --> 00:05:01.520 Artificial Analysis also shipped a speech-to-speech quality index. Twenty-seven voice models ranked on reasoning, latency and price. 26 00:05:01.520 --> 00:05:13.760 And the split is useful. One model leads conversation, a different one leads tool use. So the right pick depends on whether you're building chat or an agent that actually does things. 27 00:05:13.940 --> 00:05:27.060 Baidu open-sourced an OCR model that reads a whole document in one forward pass. A constant KV cache, 93 percent on OmniDocBench, six points over the DeepSeek baseline, at three billion total. 28 00:05:27.060 --> 00:05:42.180 That's the genuinely clever bit, swapping decoder attention for a sliding window so the cache doesn't blow up over dozens of pages. Ignore the social posts about beating 235-billion models. The paper doesn't say that. 29 00:05:42.180 --> 00:05:51.460 Oak pitched version control built for agents. Mounts a repo without a full clone, one branch per task so parallel agents don't corrupt a shared git directory. 30 00:05:51.460 --> 00:06:04.660 A real attempt at the multi-agent merge problem, which is a problem nobody had two years ago. Self-reported benchmarks, sitting at v0.96, but somebody's thinking about the right thing. 31 00:06:04.660 --> 00:06:12.820 And Armin Ronacher wrote that the harness is the new unit of work. The outer loop that supervises and re-queues an agent, not the prompt. 32 00:06:12.820 --> 00:06:23.700 No numbers in it, but he named what every agent team is converging on. The loop keeps the task alive past where the model says it's done. That's where the durable value sits now. 33 00:06:23.880 --> 00:06:32.280 LastPass says hackers stole support case data through a breach at a vendor, Klue. Second LastPass-linked incident in recent years. 34 00:06:32.280 --> 00:06:41.480 And support systems hold more sensitive context than teams assume. Your posture is only as strong as your least careful third party. 35 00:06:41.480 --> 00:06:49.160 Separately, Meta left mandatory employee keystroke logs visible companywide after a permissions error. 36 00:06:49.160 --> 00:06:59.480 That one's the lesson for everyone hoarding telemetry to train on. The more you collect, the bigger the blast radius when a single permission flips the wrong way. 37 00:06:59.660 --> 00:07:05.980 Anthropic shipped Claude Tag, an always-on teammate that ingests your company Slack to build org context. 38 00:07:05.980 --> 00:07:20.300 Sticky by design. Once Claude holds your institutional memory, switching providers means rebuilding it from scratch. And the timing's almost funny, shipping an always-on tool on the same run of capacity outages. 39 00:07:23.540 --> 00:07:26.260 Quick break — two from the desk. 40 00:07:26.260 --> 00:07:40.340 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. 41 00:07:40.340 --> 00:07:49.940 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. 42 00:07:54.210 --> 00:08:00.370 The Steam Machine launches today, topping Hacker News at 1,463 points. 43 00:08:00.370 --> 00:08:06.530 Mistral shipped OCR 4, pulling 281 points the same morning. 44 00:08:06.530 --> 00:08:13.170 a16z led a 34-million Series A in Probook, vertical AI for technician dispatch. 45 00:08:13.170 --> 00:08:25.730 And AI super PACs dropped 27 million on a single New York local race. Regulation is now a funded campaign, and the rules your stack runs under are being set in contests like that one. 46 00:08:25.730 --> 00:08:32.690 Hamel Husain and Shreya Shankar released a free twelve-talk course on evals and retrieval. 47 00:08:34.070 --> 00:08:47.030 Our call: within 90 days, at least one top-five AI coding tool ships automatic failover that reroutes off Claude on overload errors, and markets it on reliability, not price. 48 00:08:47.030 --> 00:09:01.510 We're wrong if by September 24th none of Cursor, Claude Code, Copilot, Windsurf or a Cline-tier tool ships documented failover triggered by Anthropic's overloads. That's when it settles.