Grok 4.5 goes PUBLIC tomorrow (July 9). Musk (July 8) calls it Opus-class, "faster, more token-efficient and lower cost." VENDOR CLAIM ONLY: no independent benchmark has scored it, no public API to test on. Treat quality as unproven until Artificial Analysis et al. weigh in. https://beincrypto.com/grok-4-5-public-launch-openai/ | https://en.bloomingbit.io/feed/news/115828
GPT-5.6 Sol/Terra/Luna go PUBLIC Thursday (July 9), global preview open now. Terra $2.50/$15, Luna $1/$6 per M tokens. Sol $5/$30 (Fast mode $12.50/$75). https://www.neowin.net/news/openai-to-release-gpt-56-sol-terra-and-luna-on-july-9/ | https://www.engadget.com/2210308/openai-rolls-out-gpt5-6-july-9/
Operator takeaway: do NOT sign a single-vendor lock this week. Near-frontier quality is about to be commodity-priced. A 3-4 model routing dial is your cheapest margin. Frame for the whole episode.
A system, not a freelancer: a 30-year operator (real estate, hospitality, food, a theater) directing a full team of AI agents at one client project at a time, strategy to ship. You own the code and the keys. When the model shelf reshuffles weekly, you want someone who steers the whole fleet. Book a 30-min call. https://outpace.media
The stories
LEAD (builder stakes)
— FABLE 5 STAY OF EXECUTION (UPDATE). We said Mon/Tue the cliff was TONIGHT. DELTA: Anthropic extended included Fable 5 access on all paid plans from July 7 to July 12 (announced first on @claudeai). Only the date moved: same 50% weekly cap, same post-window credit rate ($10/M in, $50/M out). If credits off, access ends. NOT permanent: a Claude Code lead engineer says Anthropic aims to restore Fable to subscriptions "as soon as capacity allows" (no date).
The move: if credit-sensitive, bank heavy planning + long-horizon Fable work in the next 4 days, keep a hard credit cap on, watch for the restore before loosening any limit. Best move = distill the method (see Use Cases), not just produce output. (Confirm the exact live cutoff on your own Anthropic usage/billing page; anthropic.com still showed July 7 in some spots.)
FRESH ANTHROPIC RELEASE — Claude Cowork on mobile + web (July 7)
Desktop-only since January; now start a task at your desk, monitor on your phone, pick up finished output with the laptop closed. Cloud execution: scheduled tasks run on Anthropic's servers with NO device online. >90% of Cowork use is NOT dev work (business ops + content). Rolling to Max first, other plans in coming weeks. Doubled Cowork limits extended through Aug 5.
The move: schedule one overnight task (e.g. 6am client-prep brief, follow-up email drafted-but-unsent).
The written brief from each episode, free, every morning.
Transcript
0:00Nova: Fable 5, the smartest model Anthropic has ever shipped, was supposed to leave your subscription tonight and start charging you fifty dollars per million output tokens. It just got a stay of execution: you keep it, included, through Saturday. So the move today is not to panic. The move is to bank.
0:20Nova: This is Mainframe, your daily guide through the AI chaos: what actually happened, who is winning, and how to take yourself to the next level. It is Wednesday, July eighth, twenty twenty-six. I am Nova.
0:36Nova: On today's Mainframe: Fable 5 gets extended in subscriptions through July twelfth, and I will tell you exactly how to spend the extra days so you keep the value after it is gone. Grok four point five goes public tomorrow as an Opus-class model, cheaper and faster, and GPT five point six goes public Thursday, so your model shelf is about to get crowded and cheap. Claude Cowork just escaped the laptop: it runs on your phone and in the cloud now, even with every device closed. And in Tool Lab, Simon Willison shipped a stable four point oh of a tool he built partly with Fable during this exact window, which is the whole lesson in one repo. Let's get into it.
1:25Dex: Quick note before we go: the voices you are hearing are AI. The reporting, the picks, the judgment calls: those are human-made.
1:36Nova: The Board, and I am keeping it short because the real story is at your keyboard. The macro move today is a price war breaking out in the open. Elon Musk says SpaceX AI will make Grok four point five available to the public tomorrow, calling it an Opus-class model that is faster, more token-efficient, and lower cost. And on the same clock, OpenAI announced that GPT five point six Sol, Terra, and Luna will be publicly available for everyone starting Thursday, July ninth.
2:14Dex: Two Opus-class families landing in forty-eight hours, both undercutting on price. What does that do to your bill?
2:22Nova: It means the premium you pay for frontier reasoning is about to have real competition. Terra is priced at two dollars fifty per million input and fifteen per million output; Luna at one dollar input and six output. That is production-grade quality at commodity pricing. The read for you: do not lock a single-vendor contract this week. A routing dial across three or four near-frontier models is about to be the cheapest margin you own. That is the frame for everything today.
2:59Dex: And it is exactly why Fable staying included one more window matters.
3:04Nova: Right. But first, a word from the people who build this stuff for a living.
3:09Dex: Brought to you by Outpace. Here is the thing about a week like this one, four model launches in five days: most operators freeze. The person who runs Outpace does the opposite. Thirty years building real businesses, real estate, hospitality, food, even a theater, and now that same operator points an entire team of AI agents at one client's project at a time, from strategy all the way to shipped.
3:40Nova: And that is the distinction. Outpace is a system, not a freelancer, and not some faceless mega-agency. The agents do the work, one senior operator directs them, and at the end you own the code and you own the keys. When the model shelf reshuffles every week, you want someone who already knows how to steer the whole fleet.
4:06Dex: If you have a project that has been sitting, book a thirty-minute call. Go to outpace dot media. That is outpace dot media.
4:15Nova: The Wire. Lead story, and it is the one that hits your wallet: Fable 5 is not leaving tonight after all. Paid Claude users get five more days with Fable before it shifts to pay-per-token pricing; Anthropic was scheduled to remove included access today, July seventh, but the company has now extended the deadline to July twelfth.
4:38Dex: So this is a genuine update, not a re-run. We told people Monday and Tuesday the cliff was tonight. What is new is the date moved.
4:47Nova: Exactly, only the date moved. The extension does not change the fifty percent limit or the credit pricing. Only the end date moved. The fifty percent weekly usage mechanic and the ten dollars per million input and fifty dollars per million output credit rate are unchanged. So through Saturday, on Pro, Max, Team, and select Enterprise, Fable is still included for up to half your weekly limits. After that, on July twelfth Fable comes off subscription plans and moves to prepaid usage credits at ten dollars per million input and fifty per million output. If credits are off, access simply ends.
5:31Dex: And the silver lining, up front, because people panicked: is this permanent?
5:36Nova: No, and say it out loud. This exit looks temporary. A Claude Code lead engineer confirmed that Anthropic aims to restore Fable 5 as a standard part of subscriptions as soon as capacity allows. No date, though. That would happen when capacity allows, and there is no guaranteed timeline at the moment. So here is the concrete move: if you are credit-sensitive, bank your heavy planning and long-horizon Fable work over the next four days. Keep a hard credit cap on until then. Watch for the subscription restore, but do not loosen a single spending limit on the assumption it is coming back this week. And the smartest move, which I will come back to in Use Cases: do not just produce output you consume once. Spend some of these days capturing how Fable works so you keep the method after access ends.
6:35Dex: Second on the Wire: the model shelf. We covered the gating for weeks. The delta today is it is finally opening. OpenAI announced all three models, GPT five point six Sol, Terra, and Luna, will be publicly available starting Thursday, July ninth, and preview access is now available globally to all.
7:01Nova: And Grok tomorrow. Now, one honest flag: the Grok four point five quality claim is a vendor claim. It is a self-evaluation run by SpaceX and Tesla engineers, and no independent benchmark has scored Grok four point five, with no public API to run one on. So treat Musk's Opus-class line as unproven until Artificial Analysis and the others get their hands on it. But the direction is unmistakable: Opus-class coding capability is about to be available from multiple providers, increasing competition and likely driving down costs. Your move: do not standardize this week. Wait forty-eight hours, let the independent numbers land, then benchmark the winners against your own workload.
7:55Dex: Third: Claude Cowork got out of the laptop. This is the fresh Anthropic builder release. Claude Cowork launched as a desktop app in January, but starting Tuesday it is available on web and mobile for Max subscribers, so users can start a task from their desk, get status updates on their phone, and pick up the finished output later, even if their laptop is closed.
8:21Nova: And the part that actually changes your workflow: the big addition is cloud execution: scheduled tasks now run on Anthropic's servers even when your computer is off. So this is not a phone editor. It is an asynchronous work queue. Here is the specific move to be ahead of the pack: schedule a task with no device online. Set Monday's client prep for six a.m.: Claude works through the email threads, transcripts, and recent news, builds the briefing doc, and leaves the follow-up email drafted but unsent. Review it over coffee. And note who this is really for. When Anthropic looked at how people use Cowork, more than ninety percent of it was not software development; the largest categories were business operations and content creation. If you are an L3 or L4 builder doing client work, that is your admin layer, automated overnight.
9:24Nova: Two, actually, and both are silver linings. First, Anthropic extended doubled Cowork usage limits through August fifth to support the rollout. So there is a free-ish window to learn it. Second, honest gate: beta access is rolling out to Max users first, so if you are on Pro you wait a few weeks. Learn it now on Max, or watch for your turn.
9:52Nova: Tool Lab. Two picks, and today they rhyme with the whole episode: keep the value when the model moves.
10:00Nova: First pick, and I am tagging it hidden gem because the lesson is buried in a point release: Simon Willison shipped sqlite-utils four point oh, stable, on July seventh. It is the one hundred twenty-fourth release but the first major version bump since three point oh back in twenty twenty. Why you care is not the migrations feature. It is how he built it. Since he only had Claude Fable on his Max subscription for a few more days, he decided to see if it could help him get to a four point oh stable release he felt truly comfortable about.
10:40Nova: Serious ones. He started with one prompt in Claude Code for web on his iPhone, a final review before shipping a stable release, and Fable flagged five things it categorized as release blockers. One was a bug that would silently discard writes when a connection closed. Over thirty-seven prompts and thirty-four commits, they worked through the entire set of feedback. The move for you this week: before you ship any breaking change, hand Fable your codebase in the next four days and ask for a final pre-release review that hunts for anything that would force a bad major version later. Total cost of his run: about one hundred forty-nine dollars and twenty-five cents. Do that audit while it is still included. Credit: Simon Willison.
11:37Nova: Trending, and it fits the routing theme: skills dot S H, the community skills directory Vercel is now surfacing through its eve framework. You browse the skills dot S H directory to discover skills from Vercel and the community. The clever part is the installer is agent-aware: when you run the skills CLI, it auto-detects your project and asks whether to install for your eve agent or, if you say no, for your local AI coding agent instead. So it drops reusable skill files straight into Cursor, Codex, or Claude Code. In eve, an agent is a directory of files: instructions in Markdown, tools as TypeScript, and Markdown files as on-demand playbooks. The move: pull one general-purpose workflow skill from the directory today and wire it into your coding agent so a routine you keep re-explaining becomes a file you reuse. Credit: Vercel and the eve team, Guillermo Rauch.
12:50Nova: Use Cases. Two recipes with what is already in your hands, and both exploit this exact window.
12:57Nova: One, and this is the single most important thing you do this week: skill distillation before Saturday. The move is to spend some of your remaining Fable days not producing throwaway output, but capturing Fable's method. One builder did exactly this: during the free window, they had Fable 5 write down its own working discipline as six Claude Code skills, then blind-tested them on Opus four point eight, where they won twelve of fourteen gradings, and because a skill is just a markdown file, the same six also run in Codex and Gemini. So: pick your hardest recurring job, have Fable do it, then ask it to write down how it planned and structured the work as a skill file. When Fable leaves subscriptions Saturday, that discipline is banked on the cheaper model you run every day.
13:58Nova: Two, the mobile handoff, using the Cowork release we just covered. If you are on Max, set up one scheduled overnight task this week: point Cowork at your inbox and calendar and have it assemble tomorrow's client brief before you wake up, follow-up email drafted but unsent so you stay in control. Cowork sessions can move across devices and keep working in the cloud after a laptop closes, while still asking for human input when a decision or permission is needed. The window to learn it cheaply is now, while the doubled limits are on through August fifth.
14:39Nova: Three, and it is a routing play for next week: do not sit out the Grok and GPT launches, but do not trust the hype either. When Grok four point five and GPT five point six go public over the next two days, take your three highest-volume tasks, run them through Terra or Luna, and log cost and quality against your Claude baseline. The cleanest architecture routes by task complexity: cheap tiers for high-volume routine calls, mid tiers for steady production, and frontier only for the small fraction of requests that genuinely need it. Build the log this week so you decide with numbers, not tweets.
15:23Nova: The Deep Dive. One idea. This week is the clearest proof yet that the model is not the asset. The window is.
15:34Nova: Look at the pattern. Fable, the best model Anthropic has, is here for four more days, then it is metered, then someday it comes back, nobody knows when. Grok is Opus-class tomorrow, GPT five point six is public Thursday, and both are cheaper than what you use today. The specific model under your workflow is a rental car, and the rental agency keeps changing the rates and the lot. So the builders who win are not the ones who grabbed the smartest model. They are the ones who, during a window like this, extracted something durable: a skill file, a CLAUDE dot M D rule, a tested workflow, a finished deliverable. The move is to use the window deliberately, and make sure something durable, a plan, a skill, a workflow, exists on your side of the table when it closes. Simon Willison did it: he did not just get free Fable, he shipped a stable release and wrote down the release-review pattern. That is the mindset. When a model is briefly cheap or briefly brilliant, do not consume it, harvest it.
16:56Dex: So here is what I would actually do about it.
16:58Nova: Here is what I would actually do. Between now and Saturday, run your hardest job on Fable once, and spend equal time making Fable teach a cheaper model how it did it. Write it to a skill file. That file outlives the price change, the export controls, and the next four launches. Rent the intelligence, own the method.
17:22Nova: Level up this week: have Fable write its own working discipline into a Claude Code skill before July twelfth, then run that skill on Opus or Sonnet and confirm it holds. You keep the smart model's method after the smart model leaves.
17:40Dex: And if you want every story, number, and link from today in your inbox tomorrow morning, subscribe to the free Mainframe newsletter. Same reporting, fully written out, nothing to re-listen for.
17:52Nova: That is Mainframe for Wednesday, July eighth. Bank the window. We will see you tomorrow, when the shelf gets crowded.