dax moves part of his team off Astra after spend doubles

dax said a portion of his team has gone back to Sol. His reasoning was not that the newer model is bad. He wrote that astra is good and can do some novel things but has downsides, and that effective spend looks doubled, which makes it tough to justify.

dax

@thdxr

X

and so far our effective spend looks doubled so tough to justify

That is one team's bill, not a benchmark. Nobody in these posts published a controlled comparison, so what exists is a spend number from a team that switched and a handful of individual impressions around it.

The impressions

Mario Zechner said Astra now writes absolutely terrible C++. Theo, who had already been complaining about what he calls Astra's spikes of stupidity, said he tried to warn people and was called mad. He also told @zeeg that a take of his was the worst of all time, said he genuinely believes @zeeg is bad at prompting if he believes it, and promised a video on it soon. So the public argument is about quality, and it is loud and unresolved.

The decision was about price

The thing dax actually acted on was cost per unit of work, not vibes about output. Kent C. Dodds pushed the same point back at him, saying he wants more models competing on cost because today's models are already pretty powerful and the real pain is in the finance department. That is the comparison teams are now running out loud, and a model that does novel things at twice the effective spend can lose it even while winning on capability.

The capacity signal

Separately, Tibo said they are pausing subscriptions to the $200 Pro plan, describing those subscriptions as the ones that put the most strain on their systems and calling the pause the smallest step that keeps access as broad as possible. All other plans and the API remain available, there is no impact to existing accounts, and he said they are working on adding more capacity as fast as they can.

For working engineers, treat those two things as separate stories that happen to share a week. The spend report is one team's experience and should be checked against your own invoices before you migrate anything. The paused tier is firmer evidence of something, namely that the heaviest subscriptions are expensive enough in compute that a lab would rather stop taking money than serve them badly. Nobody in these posts has shown whether the quality complaints and the capacity strain are connected.

Kent C. Dodds 🐨

@kentcdodds

X

there's a significant pain point in the finance department that needs solving

Mario Zechner

@badlogicgames

X

astra writes absolutely terrible C++ now.