Claude Joins Your Slack Channel. Fable 5 Comes Back Online.
Anthropic's biggest week of the year: Sonnet 5 ships at daily-use pricing, Fable 5 returns after a 19 day shutdown, and Claude Tag joins Slack. Plus a real cost breakdown for a multi-model AI council.
What matters today
Anthropic's biggest week of the year: Sonnet 5 ships at daily-use pricing, Fable 5 returns after a 19 day shutdown, and Claude Tag joins Slack. Plus a real cost breakdown for a multi-model AI council.
Key points
- Pick one real executive task this week that needs repeated multi-step reasoning.
- Pick one decision where you would normally ask two or three models separately.
- Check your current workflows for any use case where frontier-level reasoning would add value, then block out one pilot.
- Create one OpenRouter workspace with accounts for Claude, GPT, Gemini, and Grok before your next decision memo.
- Add the 4-step verify block to one template you use daily and remove vague prompts from that flow.
Welcome byte
I have not seen a week like this from Anthropic all year. Claude Sonnet 5 launched with real agentic gains at a price built for daily use, Fable 5 came back online after a 19 day export control shutdown, and Claude Tag turned Claude into a persistent Slack teammate. Sakana AI answered with a ready to use council-of-experts system, and this issue's Pro Tip shows exactly what a real multi-model setup costs with OpenRouter.
Below, I have 5 Quick Hits, 3 deep dives, a Pro Tip with a copy-ready council prompt, a Productivity Gem that adds a verify loop to any model, a Health Tip for forecasting energy from Apple Watch recovery signals, and a Kids Tip where a child can start programming a robot in a free browser simulation. Every item gives you something useful before you open the deep dive.
Quick Hits
Claude Sonnet 5 is live.
Pricing runs $2 per million input tokens and $10 per million output tokens through August 31, and Sonnet 5 is now the default model for Free and Pro plans. Read more
Fable 5 is back for everyone.
It returned today on every core Claude surface plus AWS, 19 days after a Commerce Department export control order took it and Mythos 5 offline on June 12. Read more
Claude Tag puts Claude inside Slack.
It joins as a persistent, memory-keeping teammate, in beta for Team and Enterprise plans. Read more
Sakana AI shipped Fugu.
The trained conductor model routes one request across a pool of frontier models and hands back one synthesized answer. Read more
OpenAI previewed GPT-5.6.
Sol, Terra, and Luna went to about 20 partner organizations, posting a sharp jump on a new genomics reasoning benchmark. Read more
Top AI Updates
1. Claude Sonnet 5: Real Agentic Gains at a Price Built for Daily Use
Closes in on Opus 4.8 on many tasks at Sonnet pricing, including a 63.2% SWE-bench Pro score versus Opus 4.8's 69.2%. Live on AWS Bedrock, Google Vertex AI, and Microsoft Foundry with GA this week, and now the default model for Free and Pro plans.
Why it matters: I think this is frontier-adjacent performance most Executives can already afford. Switch your default model and test one real multi-step task this week.
Action items:
- Pick one real executive task this week that needs repeated multi-step reasoning.
- Run it once in Sonnet 5 and once with the current model you used before, then compare both outputs side by side.
- Capture one metric where quality improved, such as fewer revision loops or shorter response time.
2. Sakana Fugu Ships a Ready-to-Use Council of Experts System
Sakana AI's Fugu acts like a trained conductor for multiple frontier models, routing one request through a council and returning one synthesized answer instead of making you compare tabs by hand.
Why it matters: I think this is the practical version of model diversity. Use it for decisions where disagreement is valuable, not routine drafting.
Action items:
- Pick one decision where you would normally ask two or three models separately.
- Run it through Fugu or a similar council setup and require the final answer to name disagreements.
- Compare the synthesis against your best single-model answer before trusting the extra cost.
3. Fable 5 Is Back, Fully Available, After a 19 Day Government Shutdown
A narrow jailbreak reported by Amazon researchers triggered a Commerce Department export control order on June 12. More than 80 cybersecurity CEOs signed an open letter disputing the risk. Commerce lifted the order June 30, and Fable 5 now carries no access restriction on the Claude Platform, Claude.ai, Claude Code, Claude Cowork, or AWS, while Google Cloud and Microsoft Foundry are still catching up.
Why it matters: I think model availability just became a policy decision, not just a product one. Save Fable 5 for your hardest reasoning work, not daily drafting.
Action items:
- Check your current workflows for any use case where frontier-level reasoning would add value, then block out one pilot.
- Confirm one approved account path for Fable 5 access and share it with your team.
- Define a fallback to Sonnet 5 for the same pilot so teams are not blocked by one-model risk.
Pro Tip
The Multi-Model Council Prompt That Actually Works (and What It Costs)
I like asking several AI models the same question and combining their answers, it catches blind spots a single model misses. OpenRouter (openrouter.ai) is the one account I use to reach Claude, GPT, Gemini, and Grok, so I am not juggling three separate logins. Once you have an account, try the prompt I use: "Answer using Claude Sonnet 5, GPT-5.5, and Gemini 3.1 Pro. Give a short answer from each, then combine the strongest points into one final recommendation and flag anything they disagreed on." I have found that running this across three models costs about 6 to 7 times a single chat message (roughly 6 cents versus a penny), worth it for decisions with real stakes, not routine drafting.
Action items:
- Create one OpenRouter workspace with accounts for Claude, GPT, Gemini, and Grok before your next decision memo.
- Run the same strategic question through all three outputs and keep one final answer with disagreements called out.
- Compare the dollar cost versus value and decide whether model diversity is worth the extra tokens.
Copy-ready guide The Multi-Model Council Prompt That Actually Works (and What It Costs)
Productivity Gem
Give Any AI Model a Verify Loop Before It Answers
I paste this on top of my most complex prompts: "Before you answer, work through four steps silently: understand what I actually need, sketch two possible approaches and pick the stronger one, do the work carefully, then check your own answer against my original request before you reply. Give me the final answer plus a short list of what you checked." It forces a plan-then-check step instead of a first-draft answer. Anthropic research found a 54% improvement on one benchmark from adding a verify-before-responding step like this one.
Action items:
- Add the 4-step verify block to one template you use daily and remove vague prompts from that flow.
- Ask the assistant to show assumptions and uncertainty levels before approving any recommendation.
- Build a one-page check list so a teammate can review every output before forwarding it.
Put it to work Give Any AI Model a Verify Loop Before It Answers
Health Tip
Automate a Real Energy Forecast From Apple Watch Data, Not a Sleep Score
A single Sleep Score only rates last night. I do not think it can tell you when today's energy will actually peak, because that depends on sleep debt, circadian timing, HRV, and resting heart rate. Start in Apple Health, collect wake time, sleep totals, resting heart rate, and HRV, then use this prompt: "Using my wake time, sleep totals, resting heart rate, and HRV, estimate my 3 highest-energy 90-minute windows today and 2 lower-energy windows. Base it on recovery signals, not a single sleep score. If a signal looks outside my normal range, flag it instead of guessing." Then use a 25-minute Shortcut only to block the suggested windows on your calendar.
Step-by-step guide Automate a Real Energy Forecast From Apple Watch Data, Not a Sleep Score
Kids Tip
Start Programming Reachy Mini Before You Own One (Ages 8-16)
The child is building a tiny robot behavior: make Reachy Mini turn toward a sound, then test the same idea in simulation before any hardware purchase. Start with Hugging Face's free Reachy Mini browser demo, then use the official simulation guide to run the first movement and audio notebook. The useful lesson is simple: sensors collect a signal, code decides what it means, and the robot moves. Start here: Hugging Face Reachy Mini simulation guide .
Try the activity Start Programming Reachy Mini Before You Own One (Ages 8-16)
Wrap Up
That is the week, as I saw it. Pick one item, put it into production, and reply to let us know how it went. See you next Thursday.
About the author
Pierre has spent 25+ years building growth systems across fintech, real estate, lending, campaigns, and AI workflows, with $1.5B+ in client value delivered.
Email usThree deep dives. Four useful moves. One email worth opening.
PromptHacker turns the AI firehose into practical next steps for work, health, family, and everything time keeps trying to steal.