The $2,180/Month Individual Bill That Changes Everything
Simon Willison ran the numbers on his own usage of Claude Code and OpenAI Codex using the ccusage tool. The result: $1,199.79 for Anthropic Claude Code and $980.37 for OpenAI Codex in a single month — totaling $2,180.16 in API costs for just one developer.
That's $200/month in subscriptions getting you $2,180 in actual usage. For an individual power user, that's still a great deal. But Willison realized something that most analysts missed: companies are paying dramatically more.
Enterprise Plans Got a Lot More Expensive — Fast
Between November 2025 and April 2026, both OpenAI and Anthropic quietly restructured their enterprise pricing. The old model — "seat-based pricing with bundled usage" — is gone. In its place:
- Anthropic Enterprise: $20/seat/month + full API pricing for all usage above the bundled threshold. As of April 2026, frontier models like Opus 4.7 run at approximately 1.4x the price of their predecessors.
- OpenAI Codex: As of April 2, 2026, pricing was updated to align with API token usage instead of per-message pricing. On April 23, this change was applied retroactively to all existing ChatGPT Enterprise, Edu, Health, and Gov plans.
Why the Sudden Aggression?
Both companies are planning IPOs. But the more fundamental explanation, as Willison argues, is product-market fit — specifically, with coding and general-purpose AI agents.
Tools like ChatGPT have enormous consumer reach: 900+ million weekly active users as of early 2026. But converting that into sustainable revenue is hard when only 5.6% of users are paying subscribers. At $10-20/month per user, you need 1-2 billion subscribers paying for four years just to cover $1 trillion in infrastructure costs.
Coding agents change the math entirely.
These tools burn vastly more tokens per session than a chat interface. A coding agent running in a Claude Code or Codex environment can spend $50-200 in API tokens on a single complex task. And unlike consumer ChatGPT, corporate customers see a direct, measurable ROI: one AI agent can automate hours of junior-to-mid developer work. At $200+/month per seat, that's still cheaper than one hour of a senior engineer's time in most markets.
What Changed in November 2025
The models released in November 2025 elevated AI agents from "interesting toys" to "genuinely useful daily drivers." Six months later, companies are spending real money on these tools. The Enterprise pricing shifts are a direct response to this demand signal.
The implications for builders and technical founders:
- AI agent infrastructure is a real market, not a hype cycle. The customers paying the bills are enterprises, not individual developers.
- Token efficiency matters more than ever — at $200+/month per seat, every optimization in prompt length and model selection compounds.
- The agent tooling ecosystem is still wide open — the primitives (MCP, tool use, context management) are maturing, but the tooling around monitoring, cost attribution, and multi-agent orchestration is early.
The Bottom Line
OpenAI and Anthropic are no longer burning money on discounted enterprise deals. They're charging what the market will bear — and the market, it turns out, will bear quite a lot. For technical founders building in the AI agent space, this is the strongest possible signal: the enterprise is here, the money is real, and the race to build the infrastructure layer is just getting started.