Meta shipped Muse Spark 1.1 (July 9) with its first-ever paid API. Pricing: $1.25/M input, $4.25/M output, ~25% of equivalent OpenAI/Anthropic models; $20 free credits per new account; US-only public preview at launch; cached input ~$0.15. Closed-weight (a break from Llama). Carried hard by Rowan Cheung; Meta stock moved on it.
Benchmark read: wins are agentic/tool use (MCP Atlas 88.1, JobBench 54.7 vs Opus 4.8's 48.4); trails on pure coding (SWE-Bench Pro 61.5 vs Opus 4.8's 69.2). Keep hard coding on Claude. https://handyai.substack.com/p/model-drop-muse-spark-11
Operator takeaway: near-frontier is a four-way price war (Anthropic, OpenAI, Grok, now Meta). Single-vendor lock is a margin leak. Keep a routing dial, price per finished task, bank the spread. Frames today's Fable story.
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 shelf reshuffles four times a week, you want someone steering the whole fleet. Book a 30-min call. https://outpace.media
The stories
LEAD (builder stakes)
— FABLE 5 EXTENDED AGAIN, meter now July 19 (UPDATE). We said Mon/Sat the cliff was tonight (July 12, 11:59:59 PM PT). DELTA: Anthropic extended included Fable 5 access on all paid plans (Pro/Max/Team/premium Enterprise) through July 19, 2026 at 11:59:59 PM PT — the second extension in a week — and kept Claude Code's 50% higher weekly limits through the same date. Same mechanics: up to 50% of weekly limits, drawn from the shared pool, no action needed for the included allowance. After July 20: prepaid usage credits at $10/M in, $50/M out (>2x Opus). NOT permanent; restore to subscriptions promised "as capacity allows" (no date — plan around July 19). Note the ongoing migration: Claude Code community users are moving to GPT-5.6 Sol for routine work, reserving Fable for hard tasks. Carried by Andrew Curran ("Fable isn't going anywhere"), kimmonismus; Forbes framed it as a response to OpenAI's Sol launch.
The move: don't burn the week on throwaway output — extract Fable's judgment into files you keep (see Use Cases).
The workflow runs on one skill, /brief: daily mode sorts your morning, topic mode answers a specific question, with a citation for every claim. Guided setup optionally uses your role to personalize; connects email, calendar, documents; defaults to 8:00 AM local, configurable. Set-up: in Cowork type /schedule (or open Scheduled in the sidebar). Since Cowork moved to the cloud (July 7), scheduled tasks run on Anthropic's servers even with your device offline.
The move: schedule one weekday 8am /brief tonight — open a sorted list tomorrow instead of a full inbox.
The written brief from each episode, free, every morning.
Transcript
0:00Nova: The best model on the planet was supposed to leave your subscription tonight. It didn't. Anthropic blinked again, and Claude Fable 5 just got extended, which means you have one more week to steal its brain before the meter starts.
0:15Nova: 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 Sunday, July twelfth, twenty twenty-six. I am Nova.
0:31Nova: On today's Mainframe: Fable 5 dodges the cliff a second time, the meter is now set for July nineteenth, and we will tell you exactly what to bank before it flips to credits. Claude ships a guided morning brief you can schedule in Cowork with one skill. Meta finally puts a price tag on a model and undercuts everyone by roughly seventy-five percent. And in the Tool Lab, a local router that quietly sends every prompt to the cheapest model that can handle it. Let's get into it.
1:05Dex: Quick note before we roll: the voices you are hearing are AI. The reporting, the picks, and the opinions are put together by humans. Now, the Board.
1:17Nova: One macro move that actually touches your bill this week. Meta shipped Muse Spark one point one on July ninth, and the real story is not the model, it is the meter. According to CNBC, Meta will charge one dollar twenty-five per million input tokens and four dollars twenty-five per million output tokens, with twenty dollars in free credits for every new API account, and the preview is limited to US developers at launch. That is roughly one-quarter the cost of equivalent models from OpenAI and Anthropic. Rowan Cheung is carrying it hard, and Meta's stock moved on it.
2:00Dex: The catch is it is closed-weight now. This is the company that gave Llama away, and it just joined the paid club.
2:08Nova: Right, so the takeaway for your keyboard: near-frontier is a four-way price war now, and being locked to one vendor is a margin leak. Muse Spark's wins are agentic and tool use; on pure coding it still trails Opus four point eight and GPT five point five. Keep a routing dial, price per finished task, bank the spread. That frames today's Fable story exactly.
2:37Nova: Which brings us to today's sponsor, Outpace. Here is the thing about a week like this: the shelf reshuffled four times, and if you are a solo builder you spent half of it reading changelogs instead of shipping. Outpace is built for that. It is a system, not a freelancer: one senior operator who spent thirty years building real businesses, real estate, hospitality, food, a theater, now aims a whole team of AI agents at one client's project at a time, strategy to ship. The agents do the work, the operator steers the fleet, and at the end you own the code and the keys. When the model layer changes weekly, you want someone who already swapped the config so you didn't have to. Book a thirty-minute call at outpace dot media. That is outpace dot media.
3:39Dex: The Wire. Lead is the one with your money in it.
3:43Nova: Fable 5. We told you Monday and Saturday the meter started tonight, July twelfth at eleven fifty-nine PT. Here is the delta: it didn't. Anthropic extended access to Claude Fable 5 for paid subscribers until July nineteenth, the second extension in a week; it was set to expire July seventh, then July twelfth, and now July nineteenth. Anthropic said in a support document the promotion runs through July nineteenth at eleven fifty-nine PT, and the fifty percent increase to Claude Code weekly limits was extended to the same date.
4:24Nova: Same terms as before. Starting July twentieth, per Anthropic's email to subscribers, all Fable 5 usage runs on prepaid usage credits at ten dollars per million input tokens and fifty dollars per million output tokens, and Anthropic aims to restore the model to subscriptions once capacity allows. That is more than twice the price of Opus. Andrew Curran summed it up in three words: Fable isn't going anywhere. Forbes framed it as a direct response to OpenAI's Sol launch. But two prior cutoffs were superseded at the wire, and the support article makes no forward commitment beyond July nineteenth.
5:09Dex: So the honest read is: plan around the nineteenth, not around a hoped-for restore.
5:15Nova: Exactly. And note the migration already happening: posts on the Claude Code community show developers moving to GPT five point six Sol for routine tasks while reserving Fable 5 for complex work where its output still wins. That is the correct instinct. The move is in Use Cases: don't spend the week producing throwaway output, spend it extracting Fable's judgment into files you keep after it's gone.
5:45Nova: Second Wire item, and it is a fresh Claude feature you can actually schedule. Anthropic is rolling a guided morning brief into Claude Cowork. The whole workflow runs on one skill, slash brief: it knows where your information lives, and writes back what matters with a citation for every claim, in a daily mode that sorts your morning and a topic mode that answers a specific question. Tibor Blaho spotted the guided setup: it optionally uses your role to personalize what matters, connects email, calendar, and documents, and defaults to eight AM in your local timezone.
6:26Nova: In Cowork you type slash schedule in any task, or open Scheduled in the sidebar, and set it: every weekday at eight AM, read my Gmail, issue tracker, and Slack, run slash brief in daily mode, and write a memo of what's due, what's new, and what's urgent. Since Cowork moved to the cloud last week, these scheduled tasks run on Anthropic's servers even when your device is offline. The move this week: set one scheduled brief tonight, so tomorrow you open a sorted list instead of a full inbox. Get ahead of the builders still triaging by hand.
7:08Nova: Third, fast. OpenAI spent the week doing damage control on GPT five point six token burn, and it is worth knowing if you are on Codex. kimmonismus and Theo Browne both traced it: GPT five point six's context window sits at two hundred seventy-two thousand tokens, past which cost doubles, plus over-eager subagents burned quota. OpenAI reset usage limits, added banked resets, and shipped efficiency fixes. If your Codex bill spiked this week, that was why, and it is largely patched.
7:52Nova: Hidden gem first, and it is the perfect tool for a week where the cheapest capable model changes daily. It is called llm-router, by Yali Pollak, surfaced from the awesome-claude-code list. It is a local-first router that sits under Claude Code, and Codex and Gemini CLI, and sends each prompt to the cheapest capable model, with three-layer token compression and automatic provider fallback, protecting your premium quota and cutting cost with zero config on a Claude subscription. It is credibly engineered: nineteen hundred-plus tests, an independent RouterArena benchmark placement, and an honest local-only design with no hosted proxy.
8:40Dex: Local-only matters. No third party sees your prompts.
8:44Nova: That is the whole point. You install it under Claude Code, and instead of every keystroke hitting Fable, the trivial edits drop to a cheap model automatically, and your fifty-percent Fable allowance lasts the whole week to the nineteenth. Credit Yali Pollak.
9:01Nova: Trending pick, and it is dumb-simple useful: claude-statusbar by leeguooooo, also on awesome-claude-code. It is the most complete Claude Code status line: five-hour and seven-day rate-limit usage with reset countdowns and learned end-of-window projections, context window, prompt-cache-expiry countdown, per-session cost, plus live todo, tool, and git activity, across three styles and nine themes. It is genuinely well-engineered, on PyPI with three hundred twenty-plus tests and a sub-one-percent-CPU daemon, and the clear standout of a crowded field. In a week when Fable draws from a shared pool and burns it faster, seeing your reset countdown live is the difference between pacing yourself and hitting a wall mid-session.
9:59Dex: Use Cases. Things you can run this week with what you already have.
10:04Nova: One: the Fable seven-day heist. AI advisor Allie Miller and Forbes both published pre-deadline checklists, and the logic is right. Execution can shift to cheaper models after the nineteenth; the thinking should happen while the best thinker is still included in your plan. The sharpest single move: have Fable inventory every AI subscription and API bill against what you actually used in the last ninety days and recommend consolidations, because the inventory is trivial and the judgment is hard, computing cost per finished task instead of cost per seat. Run that this week, keep the output forever.
10:51Nova: Two, the distill move we have hammered, still the highest-leverage thing you can do before the nineteenth: run your hardest recurring job on Fable, then have it write down how it planned and structured the work as a Claude Code skill file. Markdown, so it also runs on Sonnet 5, Codex, or Muse Spark later. You are buying the expensive model's judgment once and keeping it cheap forever. Source: iwoszapar.
11:22Nova: Three, a voice-drift audit while the long context is still free. Feed Fable your archive of posts and articles and ask where your published voice has drifted; the output doubles as a portable style guide cheaper models can run forever, and it works because drift detection is a long-context pattern problem: the model must hold years of material at once to notice where your current voice hedges and your earlier voice asserted.
11:55Nova: SearchVis, also an Outpace Media product, built for exactly this audience. You ship a product, and the new front page is the AI answer. When a buyer asks Claude or ChatGPT or Perplexity for the best tool in your category, the model names you or it names your competitor, and if it does not name you, you do not exist to that buyer. Mainframe tells you what shipped; SearchVis makes sure the models know what you shipped. It tracks whether the answer engines name your brand across every engine, tells you why, and gives you the exact move to publish to win the citation. Check where you stand and start free at searchvis dot outpace dot media. That is searchvis dot outpace dot media. Be the answer.
12:47Dex: Deep Dive. One idea to carry into the week.
12:50Nova: Here is the through-line nobody said out loud this week. Fable got extended not because Anthropic is generous, but because it is capacity-constrained, and Meta just proved the counter-move: price is now the weapon, not raw IQ. When the single best model is rationed and the fourth-best costs a quarter as much, "I have access to the smartest model" stopped being an edge. It repriced to roughly zero.
13:23Nova: The thing you own. The routing rule that swaps a cheap model in when Fable meters. The distilled skill file that carries Fable's judgment after Fable leaves your plan. The scheduled brief that runs whether or not you are online. Every one of those is a layer that makes the model underneath you interchangeable, and interchangeable is exactly what a client cannot buy with an API key. So here is what I would actually do about it: this week, treat Fable's extra seven days as a raw material, not a toy. Mine it for judgment, write that judgment to disk, and wire llm-router so tomorrow's cheapest model runs it for you. The lab's capacity problem is not your problem if you already own the harness.
14:18Nova: One move this week: install llm-router under Claude Code so your Fable allowance stretches to the nineteenth, then distill your hardest recurring job into one skill file before the meter starts. Cheap by default, Fable's brain on file.
14:37Dex: And subscribe to the free Mainframe newsletter. Every story, every number, every link from today, in your inbox.
14:45Nova: That is Mainframe for Sunday, July twelfth. Go steal a model's brain while it is still free. See you tomorrow.