For the past three weeks, since GPT 5.5 launched, it’s been a tale of two cities: on one side, people in finance have fallen in love with Anthropic’s growth and CFO, and remain bullish ahead of what is likely an October IPO; on the other, enthusiasm among AI engineers for Codex has risen sharply, very likely due to a combination of factors: GPT 5.5 is genuinely a strong model (in some scenarios it even reached Mythos-level performance), the launch of Codex for Everything Else, and a third factor—the spark for today’s commentary: more generous usage limits.
Claude’s pricing changes were communicated pretty well overall, but it wasn’t exactly the news harness replacement users wanted to hear: every Claude subscription now comes with a monthly API credit allowance equal in value to the subscription price. In other words, if you pay $200, you now get both: a Claude subscription with the usage limits of Anthropic’s own harnesses, such as Claude.ai and Claude Code, for “interactive usage”; and $200 in API credits for everything else, including claude-p, OpenClaw, and more, for “programmatic usage.”
If it had been framed that way from the start, people would have seen it as a very good deal:
