OpenAI sprint week 1 recap
Day 1: WIN
Sped up Sol and Astra from ~30 to ~50 tokens per second. Most people use exactly these models, so it's a clear win.
Day 2: NOT ACCEPTED
Automatic dangerous action checks, new API access tiers, meeting notes, and a decision approval interface. Honestly, all of this is pretty much useless to me. I already always work in a no-permission-prompts mode.
And solution checking should really be part of the subscription, since classifiers cost pennies.
Day 3: WIN
GPT-6 finally showed up in ChatGPT. This is genuinely awesome. Responses got faster, web search got sped up too. Definitely one of the best days.
Day 4: NOT ACCEPTED
Tailscale support was apparently already there, they just announced it as a new feature. Instant control over the agent's run is decent, but the ultra-fast Sol mode is only available on the Pro plan. Not enough.
Day 5: NOT ACCEPTED
Predicting next messages and Dots updates, both Pro only. That doesn't meet the usefulness criterion for most users. Most people aren't on Pro anyway. Features like these are needed on Plus and more affordable tiers.
Day 6: RESET
Limit resets were given. All fair, that's exactly what was promised.
Day 7: NOT ACCEPTED
The default context size was bumped to 500k tokens, even though you could already set a million manually. They announced there's no extra charge for long context, even though there never was one on subscriptions. Basically, fake improvements.
Week one stats:
2 wins
4 rejected days
1 reset day + 1 bonus reset (on day three, to mark 30M active Codex users)
This week's biggest upgrades are obviously the model speedups and GPT-6 landing in ChatGPT. Everything else is either available to too limited a group of users, or already existed before, or is hard to even call an upgrade.
Let's see what week two brings.
