OpenAI Codex costs $100–$200

Coding agents are turning into a line item. That’s the real story here. OpenAI’s current Codex documentation now says typical usage lands around $100–$200 per developer per month, which is a very different signal from the old vibe of “pay one subscription and see what happens.” The gap is that a lot of people still think of AI coding tools as basically bundled software. But OpenAI is making the cost model much more explicit now. ### What changed here? The biggest change is clarity. OpenAI’s Codex rate card now spells out that average spend is roughly $100–$200 per developer each month, with big variation based on model choice, how many instances are running, how many automations you trigger, and how often you use fast mode. It also says Codex pricing has shifted from credits per message to credits per token, which makes the meter feel a lot more like cloud compute than chat. (help.openai.com) ### Why does that number matter? Because it turns “should we use Codex?” into “how much budget do we want to burn?” A $20 or $100 subscription feels like software. A tool that can quietly add another $100–$200 per engineer starts to behave more like CI, hosting, or API spend. That changes who gets access, how teams approve it, and whether hobbyists or students can leave it running all day without thinking about it. (help.openai.com) ### Is this separate from ChatGPT plans? Mostly, yes. OpenAI’s help pages split the world into plan access and usage consumption. Plus and Pro plans give you access to higher GPT‑5.5 limits, but Codex itself uses credits under flexible pricing. In other words, the subscription gets you through the door, but the actual coding-agent workload can still create incremental cost. That distinction is easy to miss if you’re used to “unlimited” software subscriptions. (help.openai.com) ### How limited is GPT‑5.5 access? More than people assume. OpenAI’s current GPT‑5.5 help page says free users get up to 10 GPT‑5.5 messages every 5 hours, then chats fall back to a mini model. Plus and Go users get up to 160 messages every 3 hours before the same kind of downgrade kicks in. GPT‑5.5 Pro is reserved for Pro, Business, Enterprise, and Edu plans. So even outside Codex, the “best model” is already tiered and rate-limited. (help.openai.com) ### Why does fast mode matter so much? Because speed is one of the easiest ways to overspend. OpenAI explicitly calls out fast mode as a factor in monthly Codex cost. That means the expensive part is not only model intelligence — it’s also how impatient your workflow is. A team that wants instant code review, lots of background automations, and multiple parallel sessions can end up paying for responsiveness the same way companies pay extra for low-latency cloud services. (help.openai.com) ### What does this mean for small projects? Basically, you need graceful degradation. If you’re building with Codex or GPT‑5.5, you should assume users will hit limits or budgets. That means fallback models, slower modes, capped automations, and clear usage controls. The old assumption was “AI is a feature.” The newer assumption has to be “AI is a metered dependency.” Once you see it that way, product decisions get more honest. (help.openai.com) ### Is OpenAI unusual here? Not really — but OpenAI is being unusually explicit. Plenty of AI tools hide the real monthly number behind credits, tiers, and soft caps. OpenAI’s help pages are notable because they put a concrete average on Codex usage and separate plan benefits from usage-based billing. That makes the economics easier to reason about, even if the answer is a little sobering. (help.openai.com) ### Bottom line? The important shift is not just that Codex can cost $100–$200 a month. It’s that AI coding is settling into the same category as every other metered developer service — useful, powerful, and absolutely worth budgeting before you build around it. (help.openai.com)