Hey there!
Summary of this email:
- Changing reasoning effort with Fable 5.1 does not break the cache (and why we use it in low)
- Course on how to build an Event Bus with Kafka
- Café con Codely recap (watch it here)
- The joke of the week
🎚️ Changing reasoning effort with Fable 5.1 does not break the cache (and why we use it in low)
Fable 5.1 came out more than a week ago and, honestly, it has become our favorite model for coding.
It speaks plainly again (unlike Opus 5), it writes better code and it is a bit faster. But the change we liked the most is none of those. It is effort.
With previous models, changing effort in the middle of a conversation broke the cache. The whole accumulated context was processed and billed again from zero.
So the usual thing was to pick a level at the start and never touch it. If you picked high, you overpaid on the easy tasks. If you picked low, you fell short on the hard ones.
With Fable 5.1, changing effort no longer invalidates the cache. You can raise and lower the level on every message without paying for the whole context again.
This lets us use Fable in a much more optimal way. By default we work in low and we raise the effort when the task needs it.
In low, the model makes fewer tool calls, writes less preamble and goes straight to the point. For daily work it is more than enough.
When we have to design something or debug a complex bug, we raise the reasoning effort at that point and then we lower it again.
This lets us use Fable a lot within the subscription while consuming much less quota.
Tomorrow we will be live at 9 CEST talking about which models we like, analyzing tech news and more on Café con Codely.
📯 Course on how to build an Event Bus with Kafka
Last week we started publishing the course on how to build an Event Bus with Kafka.
It is a course we are putting a lot of love into, since it is something relatively new and we really like questioning the architecture.
In the part already published we focus precisely on that, on the best architecture to build an Event Bus with Kafka.
In the second part, which we will publish next week, we focus on implementation and on retries/dead letter systems.
We hope you like it! 😊
☕ Café con Codely recap
This is the top 3 of headlines we discussed in the latest Café:
- GPT-6 Astra against Fable 5.1: two turbulent launches in one week and no model arrives in one piece. Astra is trained to use the computer, but the rollout is so slow that almost nobody has it and OpenAI compensates each day without access with an accumulated reset.
- The Evernote legacy: 750 machines provisioned by hand and a Java 11 monolith. The CTO of Bending Spoons explains what they found when they bought it in 2023, and their latest acquisition, on August 4, is Airtable.
- Manifest V2: Google deletes uBlock Origin and friends from the Chrome Web Store and Brave hosts four extensions. Google earns between 30 and 60 dollars per user per year from ads, and without an ad blocker that revenue can grow by 30 or 40%.
You have the full Café recap with links to each story at the moment we discussed it.
Tomorrow we will be live at 9 CEST discussing this week's news on Café con Codely. On our YouTube, Twitch and X. 🙌
And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:
I was going to tell you a joke about DNS, but it would take you about 24 hours to get it. 😂 😂 😂
Cheers!