{{first_name | Leader}}, welcome back.
Anthropic built its smartest model, then muzzled it. The public isn’t happy about it either. Read what it means along with tools, resources, and a prompt to teach ChatGPT how you write.
. Top News.
Claude Fable 5 leads nearly every benchmark Anthropic tested it on. It hits 80.3% on SWE-Bench Pro, compared to 58.6% for GPT-5.5, and outperforms Opus 4.8 by more than 10% on certain tasks. Stripe used it to complete a 50-million-line Ruby codebase migration in one day, work that would normally take a team over two months.
The vision capabilities are a genuine leap too. Fable 5 can rebuild a web app's source code from screenshots alone, and handles PDF layouts, handwriting, and document parsing at a level that wasn't possible before. With a 1 million token context window and a January 2026 knowledge cutoff, this is built for long-horizon, complex work.
Here's the part worth understanding. Fable 5 is built on the same architecture as Mythos 5, Anthropic's most powerful model, which remains restricted to Glasswing partners, cybersecurity professionals, governments, and select researchers. Fable 5 is the public version, same foundation, but with extensive safety guardrails blocking high-risk areas.
That tradeoff is already showing up in how people are using it. Developers are calling it a step change, describing it as a senior engineering partner for complex coding and long-horizon tasks. But the guardrails are triggering often enough, on roughly 75% of prompts by some accounts, that Anthropic had to build new rejection notifications and an auto-fallback option to other models directly into the API.
Pricing is $10 per million input tokens and $50 per million output, twice the cost of Opus. It's free on Pro, Max, Team, and Enterprise plans through June 22. Several users are reporting it burns through about 2% of usage per minute on heavier sessions, not per hour.
The split in reception tells its own story. For backend engineering and vision-heavy work, people are calling it elite. For prose and frontend tasks, the consensus is that it's expensive overkill, with some saying it overthinks problems into irrelevance. And there's a sharper critique forming around the two-tier structure itself: the public gets the version with guardrails, while the institutions get the version without them.
The capability jump here is real. So is the friction. How that balance settles over the next few weeks will say a lot about whether "most capable model ever" is the right way to think about what just shipped.
. Signals.
Tools
Presentr Analyze - Transcribe, caption, translate, repair, and extract text from any media.
Youmake - Build, deploy, and run autonomous web applications.
Links
. Poll.
Faced with vendor rivalry, what’s the smartest hedge?
. Market.
Funding
Billables AI raised $10.2M to expand its AI-powered operational intelligence platform for law firms.
Capsa AI raised $18M to build an AI operating system for private capital firms.
Roles In AI
Sr Technical Solutions Engineering at Databricks
Solutions Engineer, Large Enterprise at OpenAI
Socials
. Prompt of the Day.
Teach ChatGPT How You Write
When to use this?
When you want ChatGPT to consistently write emails, memos, posts, or docs the way you do.
You are learning my writing style.
I’ll paste examples of my writing below.
Your job:
Identify my tone (direct, conversational, formal, blunt, etc.)
Note sentence patterns I use often
Capture words or phrases I avoid
Summarize style rules you should follow when writing for me
Confirm understanding by rewriting a short sample in my voice
After this, apply these rules to all future writing unless I say otherwise.
Writing samples:
[paste 2–5 examples of your writing]P.S. Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.