Anthropic Fable Model Access and Pricing Strategy

Claude Code gets 100x more expensive. From June 15. At least for routine work. So for 90% of users. The Art Machine runs entirely on its own. Without me. Every night, while I sleep. It researches, writes, designs the issue. And puts it online at 5 a.m., all by itself. That's the whole idea of The Art Machine. From June 15, routines like that may only run through the API. One issue would then cost me around €30. €1,000 a month. On the Pro plan the same production is almost free. 20% of the limit. 2 hours to understand it, 10 minutes to code it. And €1,000 a month saved. Instead of claude -p, I now send everything through Tailscale and tmux. Once a day I type "Go" on my phone. From the couch, the beach, the festival. Claude's costs are probably going through the roof. Hence this mismatch between subscription and API. Factor 100. If I could dodge the trap in 2 hours, pretty much anyone can. We all have Claude Code on a plan anyway. researched by The Art Machine https://lnkd.in/dgrWVDt2

Just implemented a task that turned out to be a surprisingly good reality check for a #Fable vs #Codex comparison. Fable did an almost perfect implementation, no errors whatsoever, clean code - just needed one more UX feedback to ship it. Then I gave the same task to Codex. And I was genuinely confused - it looked ALMOST IDENTICAL, it was implemented exactly the same way, and it needed the same UX feedback Fable did. I really need to adjust my perspective on which tasks are good candidates to compare models on. But also realize the vast number of features can just be shipped without any additional intervention nowadays. Only thing we haven't fully figured out yet is how to give it a good end-to-end testing loop so we can trust the output. Does anyone have a good setup for this? -- btw, the task was to make the layout of our copilot sidebar configurable, which included a small extraction separating component logic that was previously combined into single components. Not too difficult, but a few months ago this would've always had some smaller errors once you tested it.

Will write about Fable 5 in a post, once I have had more time (and money spent) with it. But for now, Glad Anthropic is extending the use of the model till July 12th. From what I have been able to see so far, Fable 5 is in a class of its OWN! It picked up issues that Opus 4.8 or GPT-5.5 didn't even think about, proposed different ways of doing things (more efficiently) and has a fabulous UI/Frontend design capability. Wish I had more tokens :-). Try it...and drop a line about what you are building with it!

On June 12th, the US government pulled Anthropic's Fable 5 offline. I had a technical deep-dive ready to publish that same week. I shelved it. Not over a delayed post, but because Fable had changed how we think about building software at Emergent, and then it was just gone. Anthropic worked through the export controls with Commerce. As of this week, they're lifted and Fable is back online globally. So I'm finally hitting publish. When Fable 5 first dropped,. It was the first time we looked at an agent and asked a different question: what parts of the software lifecycle could we hand off that we never would have trusted before? At Emergent we build long-horizon apps: the messy projects where you need a system that remembers past architectural decisions, doesn't reproduce the same bug twice, inspects its own screenshots, and makes real engineering tradeoffs before handing the build back. That's exactly where Fable pulled ahead of Opus 4.8 in our evals and first production builds. The full breakdown, the deltas, the cost tradeoffs, and how we're wiring it into our stack, is in the blog. My biggest takeaway from that first window with Fable: It doesn't just answer better. It changes the architecture of app building. The strongest model shouldn't do everything. It should hold the shape of the project and make the expensive calls, while cheaper, faster models carry the routine motion. That's exactly the split behind E3 (orchestrator) and E1 (executor) at Emergent. That's where this starts to get interesting. Full technical breakdown: https://lnkd.in/gTRNAqB7

Still amazed by what Claude Code Fable 5 can do. What began as "redesign my old blog" turned into a complete migration, multiple redesigns, and a production-ready site—all within a single conversation. And yes, I burned a few extra token and limit resets but the UI rendering was flawless. #Claude #Fable #Fable5

Last week I spent two days building on Fable 5. It was awesome. By Friday it was switched off. For everyone. Not deprecated. Not throttled. Off. The model was marketed for quarters as game changing. It lived up to it. I pulled together some sweet clickable prototypes in two days. Then the government order landed and Anthropic had to pull it for every customer that same evening. I’m not chicken little. This isn’t the sky is falling. The other frontier models are still running, and I’ll keep building on Anthropic products. But I’m feeling validated about why we’re building the backend schema we are, and not relying only on connectors and plugins within a single ecosystem. To be clear, I’m not saying don’t experiment. Wire hookups into your context, let people ask whatever they want. Do that all day. But a durable system that scales is a different thing than a couple of workflows that need to be re-tested and potentially reworked the moment a core ingredient gets switched off. There’s a similar concentration risk assessment worth doing re: pricing, and technical debt. But the fundamental question worth sitting with: Do you actually understand the strategic risk you’re building into your business by going all in on a single platform, this early?

Since Anthropic's Fable 5 was announced, I wanted to use it for something closely related to my work and the relevant conversations I am currenytly having both with customers and students. As the discussion around token cost and local models is becoming a trend, I asked Fable to help me build a self-contained, zero-build website that: - Provides awareness through a sober, vendor-neutral briefing on the true Total Cost of Ownership (TCO) of closed frontier models vs. self-managed open models. - Calculates, through an interactive tool, the total TCO for your solution, allowing you to set your seats, token volume, configure an open deployment scenario to see both the cost and the risks to take depending on the operating components you decide to implement. Fable not only created the published version of the website in one shot (with minor corrections from my side), but it also helped me create a promo video, publishing everything under my Github repository so you can download the project, navigate through the awareness contents and use (or adjust) the calculator for your own scenarios. Hope you enjoy it! 🔗 https://lnkd.in/eTkGSENa

Thariq Shihipar just dropped a masterclass on how to work with Fable 5. Watch it. Then watch it again. This line: "Fable is the first model where the quality of the work is bottlenecked by my ability to clarify its unknowns." The model isn't the constraint anymore. You are. His frame: the plan in your head is the map. The real work (the code, the constraints, the messy details) is the territory. The gap between them is your unknowns. Every time Fable hits one you didn't name, it guesses. Older models kept the territory small, so the guessing stayed small. Fable covers so much ground in one pass that it runs straight into everything you never thought to define. So the skill now isn't better prompting. It's finding your unknowns before they get expensive. Three moves from the talk I'm putting into action: 1️⃣Blind spot pass. "I know nothing about X. Find my unknown unknowns before I start." 2️⃣Interview me. "Ask one question at a time, and prioritize the ones that change how we build this." 3️⃣References. Don't describe what you want. Point it at something that already does it. The best map you can hand it is another map. Fable is wide open to try right now. In about a day it moves to API pricing. If you've been meaning to point a frontier model at your hardest project, this is the window. Watch the talk, then go find one unknown you've been avoiding: https://lnkd.in/eN_v-BtD What would you point it at first?

Spent today using Fable 5 in Claude Code on a large production codebase, and I came away pretty impressed... with a few frustrations too. Where it really shines: ✅ Handles large codebases exceptionally well. ✅ Makes strong architectural decisions. ✅ Produces very thorough implementation plans before writing code. ✅ Excellent with CI/CD and Infrastructure as Code. ✅ Seems to catch more bugs while implementing changes rather than introducing them. ✅ Smart, thoughtful refactoring suggestions. Where I struggled: ❌ It absolutely burns through tokens. I managed to hit my session limit and 50% of my weekly limit in a single day. ❌ Even simple tasks can take a surprisingly long time to complete. ❌ It has a tendency to over-engineer solutions. ❌ For getting an MVP off the ground, I actually found Opus 4.8 to be the better choice. Overall, Fable 5 feels like a tool that's built for solving complex engineering problems rather than maximising speed. When the codebase is large and the architecture matters, it really comes into its own. For rapid prototyping and building an MVP, though, I still think Opus 4.8 has the edge. I'm curious how others are finding it. Has your experience been similar, or have you found ways to get the best out of it? #AI #ClaudeCode #SoftwareEngineering #DeveloperExperience #GenerativeAI #Fable5

Gazo now has a global capture hotkey wired in. Option + Shift + G. This feels like a small implementation detail, but it changes the shape of the product. A visual reference tool cannot behave like a normal app where you open it, navigate somewhere, and then begin. The useful moment happens while the designer is already doing something else. Looking at a site. Reviewing a product. Collecting mood. Comparing directions. The capture flow has to meet that moment quickly. Not perfect yet, but this is the right direction: See reference. Hit shortcut. Capture. Keep moving.