Why is GPT-6 Astra slow in long conversations?

Astra is a new model with optimization issues in long chats. Nothing wrong on our side — for speed use GPT-4.1, for balance GPT-5.6 Sol.

GPT-6 Astra is new here and got a lot of attention from day one — thank you all. Lately people ask "why is Astra slow in long conversations?" Let's explain openly.

Why does it slow down?

Astra is a brand-new model. It is fast on short prompts (2–4 seconds), but as the conversation grows — especially agent work with 10+ messages and tools — response times get noticeably longer. In our measurements, a 12-message agent history takes minutes on Astra while the same history returns in seconds on other models.

This is an optimization issue typical for a fresh model release; it improves as the model side matures. So there is nothing wrong on our side — our infrastructure is healthy.

What do we recommend right now?

• Need speed → GPT-4.1: stable in 2–3 seconds, short or long. Currently the most comfortable option for agent work.
• Need balance → GPT-5.6 Sol: good quality/speed balance, clearly faster than Astra in long chats.
• Keep trying Astra: use it for short chats. If it stalls on long history, open a new conversation or compact (summarize) the history and continue.

What did we do?

We set a 40-second fast limit for Astra requests: a stuck request no longer holds its slot for ages — you get a clear error with an alternative model suggestion instead. Other models are unaffected.

We will announce updates here. Happy building!