Simon Willison, programmer and AI commentator, argues both Anthropic and OpenAI hit product-market fit in April. New frontier models repriced 1.4x-2x. Enterprise customers now locked at API rates, not workday discounts. Coding agents are the killer app.
April 2026 was the pricing pivot. GPT-5.5 priced 2x GPT-5.4. Opus 4.7 priced 1.4x Opus 4.6. Both labs flipped enterprise plans to API metering.
Willison ran ccusage on his own machine. $2,180 of API tokens for $200 in subscriptions. Power users burn $1,000 per vendor per month. Anthropic's enterprise switched to $20 per seat plus API in November 2025, now surfacing in renewals. OpenAI Codex went all-API on April 2. Uber maxed its 2026 AI budget by March, mostly on Claude Code. Microsoft canceled Claude Code seats around its fiscal year end.
For PMs at AI-adopting companies: coding-agent line items will hit $200-1,000 per seat per month by Q4. For execs: the free-trial era ended. For procurement: when your team uses agents heavily, expect to suck air through your teeth before signing.
⚡ Why this matters
- Pricing power is the cleanest PMF signal. Both labs got it in April.
- Coding agents are now the daily driver for the highest-paid workers on Earth.
- ChatGPT had 900M weekly users but only 50M paid. Coding-agent enterprise pricing breaks that ceiling.
🔍 What happened
- May 27, 2026. Simon Willison publishes "I think Anthropic and OpenAI have found product-market fit".
- GPT-5.5 (April 23 release) is 2x the API price of GPT-5.4.
- Opus 4.7 (April 16) is ~1.4x the price of Opus 4.6 (new tokenizer adjusted).
- Anthropic Enterprise quietly changed to $20/seat/month + API pricing back in November 2025 (per Information, surfaced April 14).
- OpenAI Codex moved to API token pricing April 2 (Plus/Pro/Business) and April 23 (Enterprise/Edu/Health/Gov).
- Willison's personal usage: $1,199.79 Claude Code + $980.37 Codex per month in API equivalent for $200 in subscriptions.
- Anthropic 26.9% of open jobs are enterprise sales. OpenAI 32.6%. Both ramping go-to-market hard.
- Uber CTO Praveen Neppalli Naga: "maxed out full year AI budget just a few months into 2026", mostly Claude Code.
- Anthropic projecting $10.9B Q2 revenue, potentially first profitable quarter.
💬 Smart takes
- Willison: "Coding agents really did change everything. These are tools which burn vastly more tokens, but are also quickly becoming daily drivers for the work carried out by extremely well-compensated professionals."
- Willison on pricing: "Your customer should suck air through their teeth and then say yes. Uber's budget overrun and Microsoft's seat cancellations look like that effect playing out in practice."
- Anthropic to SpaceX customers letter: the $1.25B/month deal lets us "increase our usage limits for Claude Code and the Claude API." Heavily implies Colossus is being used for inference.
- Skeptic: April pricing was opportunistic. If a frontier model from DeepSeek or Qwen lands at 30% of GPT-5.5 price with 90% of capability, this PMF unwinds fast.
🧭 Where this goes
- Anthropic and OpenAI both file public S-1s before Q3. Audited revenue figures finally land.
- Cursor and GitHub Copilot lose API customers as Anthropic pushes Claude Code direct.
- Enterprise procurement teams set per-seat API spend caps with hard kill-switches by Q3.
- By 2027 a "coding agent budget benchmark" emerges - $400-800/dev/month is the new normal.
🎯 Implication
- For PMs at AI-adopting companies: assume coding-agent budgets land at $200-1,000 per seat per month by Q4. Build internal cost reporting now.
- For execs procuring AI: ask vendors for API rate cards, not workday allowances. The discount era is over.
- For SaaS vendors with AI embedded: your COGS just doubled with the April price changes. Reprice or eat margin.