5 Quick Links for Devs: Week 6, 2026

Anthropic Releases Claude Opus 4.6

The battle of the frontier models enters a new round, with Opus 4.6 being released with a huge swathe of improvements. Anthropic is leaning into coding & work improvements, and a much larger context length of 1M tokens. I will be testing it out this weekend!

Open AI Introduces GPT-5.3-Codex

Not to be outdone, around 30 mins after Anthropic's announcement Open AI released their new version of GPT Codex (a coding optimized version of GPT). I'm super excited about this one because over the past month of using both Opus and GPT-Codex I found GPT to be the superior coder, with excellent planning and execution skills, and much more sophisticated style of engineering. They've doubled the usage limits as well so it could be a good time to try it out!

Related: Open AI Codex App

It’s 2026, Just Use Postgres

With the massive increase of the abilities of coding models, over the past month I've also found myself using Postgres as a default database instead of MongoDB, which I used to always reach for when I wanted to spin something up really quickly. 
The reason that I tended to use MongoDB (apart from the fact I used to work for a company that made a GUI for it), is that being a NoSQL tool made it very easy to just get started without having to define strict schemas and properly designed relationships. That makes sense for the times when you want a very robust application, but often wasn't suitable for a weekend project. 
Now, however, with the frontier coding models able to spin up properly designed schemas, I've dropped MongoDB and I'm mainly using Postggres, usually with Supabase or Render when I deploy! Raja Rao makes a good case that Postgres is all you need!

My experience with vibe coding

Gabriella Gonzalez takes Claude for a test drive (as a Haskell developer) and finds that while it does come up with a solution to a pretty gnarly bug, the solution itself is not sound from the perspective of a very experience developer.

I think it's good to hear different perspectives on people's experiences with coding models, and also to reinforce the understanding that the output still has to be carefully evaluated, both in its own right and also in the context of the current project. For example, Gabriella points out "Claude's solution is not reproducible at all on other machines. In fact, it's not even necessarily reproducible on my own machine: Claude's solution will eventually stop working on my machine the first time I run a Nix garbage collection".

The API Tooling Crisis: Why developers are abandoning Postman and its clones?

The enshittification of software continues, this time in the realm of API tooling. One time crowd favourite Postman has fallen from grace, locking people out of their accounts and reducing functionality in the pursuit of profitability. Insomnia too seems to have fallen by the wayside, while Thunder Client is described as executing a 'bait-and-switch' on users. Post writer efpasia proposes an 'ideal solution' basically as a prompt, so I might have a go and see what GPT-5.3-Codex can do with it this weekend..