Day one of the tally

Day 1 came in at 22:50 India time on Monday night. "We have optimized the default speed to be ~50% faster across GPT-6 Astra and GPT-6.1 Sol through the subscription across all our products and partners using Sign in With ChatGPT." The partner list names OpenCode, Pi, Amp and Devin. Then a second post, half a minute later: 50 tokens per second instead of 30. Tally: ship.

Two things I like. It is a number, so it can be measured by anyone with a stopwatch. And it reaches third-party tools through a sign-in, not only OpenAI's own apps. That is the "relevant for most users" bar from the pledge, met on the first try. The post has about fifteen thousand likes. People wanted speed.

What it does not touch is the other number people argue about, usage per dollar. That same evening synthwavedd posted a comparison of how much each lab's subscription gives you against the equivalent API price. His read is that Anthropic's plans come out at over five times the value on mid-tier models, with Astra giving very slightly more raw usage than Anthropic's matching model at the top plans. Kimmonismus passed it on, attributed it to SemiAnalysis, and added that OpenAI had halved the limits on its $200 plan. I could not confirm that last claim from a primary source, so I am treating it as a claim and not a fact. The comparison itself is third-party math, not a lab statement.

The point for the tally is that speed is the easy kind of improvement. It is one setting, it lands in two hours, and nobody has to be asked to give anything up. The harder days are the ones where "clear improvement" means more usage, and there people are already holding a ruler. I will keep marking each day by what it moved, and I am watching for what the tweet for Day 2 names. As of this morning there is not one yet.

Day thirty-nine. One ship, no resets.