Hey there!
Summary of this email:
- Claude Fable 5 vs GPT-5.6 Sol
- What happens if my message queue goes down (+ Inbox and Outbox design patterns course available on the Standard plan)
- Grok 4.5 is a GREAT model that is NOT at Fable's level
- The Café con Codely infographic
- The joke of the week (it's got a lot of cache!)
🥊 Claude Fable 5 vs GPT-5.6 Sol
Claude Fable 5 access with the subscription had an expiration date: July 7.
But just hours before it expired, Anthropic extended it until this Sunday the 12th.
Why would you extend access to the model that costs you the most to run? Our theory: OpenAI.
Anthropic knew that GPT-5.6: Sol, Terra and Luna were launching today, Thursday.
And they knew that if Fable disappeared right before, this week everyone would only be talking about how good Sol is.
By keeping Fable available a few more days, we can all compare. And see that Fable, despite being more expensive, is probably still better in quality.
Conspiracy theories aside, what's clear is that both models have raised the bar: they're the new state of the art, above Opus 4.8 and GPT-5.5.
And they needed it.
Open source models like GLM-5.2 were already reaching Opus level. With this new generation, both OpenAI and Anthropic once again have frontier models that open source can't match.
That said, as of today, at the speed Chinese open source models are moving, we'll probably have something similar by the end of the year.
Keep in mind that this new generation comes with new-generation pricing: Fable at $10/$50 per million tokens and Sol at $5/$30. If we want to optimize costs, we can use them mainly for planning.
We're really liking Fable ourselves. And as soon as we get access to GPT-5.6 Sol, we'll be all over it to tell you about it.
We'll talk about all of this in much more detail tomorrow on Café con Codely.
📬 Design patterns course: Inbox and Outbox
What happens if you register a user and, right when publishing the event, Rabbit is down? Do you roll back and return a 500 error? Do you assume you've lost the event?
The Outbox and Inbox patterns solve exactly this: persisting to the database before publishing (or when consuming) a message, to make sure no event gets lost and to avoid consuming duplicates.
But they have a problem: adding them "just in case" is overengineering. When to apply them, when not to, and how to implement them with their table, publisher and consumer — that's what we cover in the Inbox and Outbox design patterns course, available on the Standard plan. 😊
🗞️ Grok 4.5 is a GREAT model, but it's not at Fable's level
The news to highlight this week: we can't trust benchmarks.
Grok 4.5 came out this week, and on CursorBench it shows up at Fable 5 level.
But Elon Musk himself placed it at Opus 4.7 level. Although faster, that is.
So how can it score that high? Cursor explains it in their own announcement post, hidden in a footnote:
Grok 4.5 has an advantage on CursorBench because an earlier snapshot of the Cursor codebase was accidentally included in training.
In other words: the model had seen Cursor's codebase while training. Contaminated benchmark.
They say that data has already been removed for future models and that they're working on an update to CursorBench. In the meantime, you know: take benchmarks with a pinch of salt.
If you want to try Grok 4.5, it's been available in the US since yesterday, but it won't reach the EU until mid-July (again…).
We'll also talk about this tomorrow at the Café.
☕️ The Café con Codely infographic
We're preparing some changes to the newsletter, and one of them has to do with integrating it more with Café con Codely.
This week we're attaching the summary of what we covered in the last Café, but we'll most likely have more news about this next week.
See you tomorrow at 9 CEST 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:
Why is Redis the most respected database? Because it has a lot of cache! 😂 😂 😂
Cheers!