Rumors are circulating that Anthropic is on the cusp of its first profitable quarter, while stories abound of companies surprised by their escalating LLM bills from staff usage. This suggests that both OpenAI and Anthropic have successfully found product-market fit.
Indeed, enterprise customers are now predominantly paying API prices. As a reference, a moderate user subscribing to Anthropic's $100/month Max plan and OpenAI's $100/month Pro plan found their 30-day API token usage, if paid directly, would amount to $1,199.79 for Anthropic Claude Code and $980.37 for OpenAI Codex, totaling $2,180.16. This is a significant difference compared to the $200 subscription cost. Previously, it was assumed that companies extensively using agents received similar discounts, an assumption that proved to be incorrect.
Anthropic made a critical shift in its Enterprise plan sometime in the last six months. What was initially described as “Claude seats include enough usage for a typical workday” (circa August 2025) was changed to a $20/seat/month charge plus API pricing for usage. This change, reportedly effective November 2025, was cited by an Anthropic spokesperson in an April 14, 2026, report by The Information. Existing customers are discovering this new structure upon contract renewal.
OpenAI implemented a similar pricing adjustment in April. As of April 2, 2026, Codex pricing was updated to align with API token usage, moving away from per-message pricing. This applied to new and existing Plus, Pro, ChatGPT Business, and new ChatGPT Enterprise plans. By April 23, 2026, this update extended to all existing ChatGPT Enterprise plans, including Edu, Health, Gov, and ChatGPT for Teachers. While OpenAI quotes prices in “credits,” these credit costs appear to be an exact match for the API token costs of their respective models. Essentially, by April 2026, the enterprise cost for both OpenAI Codex and Anthropic Claude Code/Cowork became equivalent to their listed API prices.
April also saw both leading model companies release new frontier models at higher API prices. GPT-5.5, released on April 23rd, is priced at twice the API cost of GPT-5.4. Opus 4.7, launched on April 16th, is approximately 1.4 times the price of Opus 4.6 when accounting for its new tokenizer. This indicates a strategic move by both companies to capitalize on increased demand and perceived value for their advanced models, further solidifying the usage-based revenue model.
[AgentUpdate Depth Analysis]
This critical shift by Anthropic and OpenAI towards API-based pricing for enterprise users, coupled with the release of more expensive, powerful models, signifies a profound maturation of the AI Agent ecosystem. It validates that enterprises are no longer merely experimenting; they are deriving significant, measurable value from AI agents, compelling them to pay for actual usage rather than flat rates. This paradigm shift will force AI agent developers to prioritize cost optimization, refine prompt engineering, and implement intelligent token management strategies. It encourages the development of more efficient, specialized, and domain-specific agents, moving beyond general-purpose models. Frameworks like LangChain or Auto-GPT will need to evolve to integrate sophisticated cost analytics and optimization tools, empowering businesses to manage their AI expenditures effectively. Ultimately, this move establishes a more economically viable and performance-driven agent market, where efficiency and demonstrable ROI become paramount, accelerating the integration of autonomous AI agents into core business operations.
