OpenAI's Token Diet: How Codex Ditched $25/Month for Pure Usage Billing
Everyone says usage-based pricing is the future. But when it comes to developer tools, most teams prefer the devil they know—predictable monthly bills that don't spike when junior devs go prompt-crazy.
OpenAI clearly didn't get that memo.
The Numbers That Matter
Codex now offers pay-as-you-go pricing for teams, completely eliminating fixed monthly fees for developers who only need intermittent access. The token consumption model covers input, cached input, and output metrics with zero traditional rate limits.
Meanwhile, ChatGPT Business got a permanent price cut from $25 to $20 annually per seat. Early adopters get up to $500 in migration credits to sweeten the deal.
<> "We're making it easier to just build things. Starting today, teams on ChatGPT Business and Enterprise can add Codex-only seats to their workspaces with pay-as-you-go pricing, giving full access to Codex without a fixed seat fee."/>
This sounds revolutionary until you realize what OpenAI isn't telling you.
The Math Nobody Wants to Do
Here's the trade-off nobody's talking about: heavy Codex API usage can average higher costs than GitHub Copilot's flat monthly rate. Copilot charges predictably. Codex scales infinitely but bills unpredictably.
For teams with variable usage? This is genius. For teams running AI-assisted development 40+ hours per week? This could get expensive fast.
The current pricing structure tells the real story:
- ChatGPT Plus: $20/month with 33-168 messages per 5-hour window
- ChatGPT Pro: $200/month with 223-1,120 messages per 5-hour window
- ChatGPT Business: $30/user/month with team controls
- Enterprise: Custom pricing (translation: "Call us")
The Elephant in the Room
This isn't about developer happiness. It's about competitive positioning against GitHub Copilot.
Copilot's flat-rate model gives CTOs budget predictability. You know exactly what 100 developers will cost next quarter. OpenAI's token model? Good luck explaining that variance to your CFO when the monthly bill jumps from $2k to $8k because your team shipped a major feature.
But here's where it gets interesting: OpenAI removed rate limits entirely for token-based billing. That means unrestricted multi-agent scaling without purchasing rigid per-seat licenses. For AI-native startups building code generation pipelines, this is massive.
What This Really Means
OpenAI is betting that development teams care more about scaling flexibility than cost predictability. They're probably right for startups and experimental teams.
They're probably wrong for enterprise teams who need to forecast budgets six months ahead.
The Codex-only seats approach is clever—it lets teams test AI coding assistance without committing to full workspace licensing. Add the GPT-5.1-Codex-Mini model option when approaching usage limits, and you've got a sophisticated cost optimization strategy.
The Real Winner Here
It's not developers or enterprises. It's OpenAI's revenue predictability.
Usage-based pricing historically generates higher lifetime value than flat subscriptions. Teams start small, get hooked, then usage creeps up. Before you know it, you're paying more per month than the old fixed model—but it happened gradually enough that nobody noticed.
Smart? Absolutely.
Transparent? Not really.
The SOC 2 Type 2 compliance and promise that business data won't be used for training addresses enterprise security concerns. But the pricing complexity creates new procurement friction.
Bottom line: This pricing model will accelerate adoption among startups and hurt predictability for enterprises. Choose accordingly.
