OpenAI's Codex plan lost 10x value. Claude is the better bet now.

The value collapse. OpenAI's $200 Codex plan once delivered up to $12,000 of monthly inference. Recent changes have cut that to roughly $2,000, a 10x drop. The creator of T3 Code walked his own usage data, showing $573 of value from a full week of heavy use. This is not a perception problem. The numbers are materially worse.

Why it happened. The subsidization arms race between OpenAI and Anthropic became unsustainable. Both labs priced subscriptions as loss leaders to capture developer mindshare, but model costs rose and usage patterns shifted from occasional prompts to intensive agent-driven workflows. OpenAI had to tighten limits or face runaway compute costs.

The hidden tradeoff. Subscribing to use GPT-6.1 Soul instead of Astra actually stretches value further. Soul is dramatically cheaper per token and per task. On Terminal Bench, Soul scored nearly as high as Opus 5.5 at one-thirteenth the cost per task. The plan still delivers strong practical output, provided you avoid the expensive models.

Where Claude wins. The Claude $200 plan offers Opus 5.5 usage that remains effectively unlimited for heavy coding workflows. Anthropic restricts model access more strictly, but for shipping production code, the value is superior. The speaker runs six Claude accounts and drains them regularly.

The $500 plan blunder. OpenAI launched a $500 tier with access to ultrafast inference on Astra, but ultrafast burns through the weekly limit in roughly two hours of generation. The speaker calculated $1,000 of ultrafast value per week versus the $12,000 the old $200 plan offered. This was an optical disaster layered on top of a real cut.

What to watch. The competitive pricing pressure is collapsing margins. Labs are moving from 95% margins toward 50% on certain models. Over the next year, subscription subsidization will continue shrinking, but the raw work output per dollar will still improve because the models themselves are becoming more efficient. The drama is real, but the practical trend is positive.