From Flat Fees to Tokenomics: A New Cost Center
The great AI price collapse refers to the shift from flat-rate AI tools to usage-based pricing that bills enterprises per token, turning every interaction with models like coding assistants, chatbots, and agents into a metered cost line that can surge with heavy workloads and unpredictable demand. GitHub Copilot’s move on June 1 to usage-based billing marked the end of the flat-rate era for many developers. Annual plans are being retired and GitHub AI Credits are now burned according to tokens consumed, priced at the listed API rates for each model. One AI Credit equals USD 0.01 (approx. RM0.05), with allowances from 1,500 to 20,000 credits depending on plan. The same pattern is spreading across tools like Cursor, Windsurf/Devin, and Anthropic’s API, pushing enterprises to treat “AI token costs” as carefully as cloud compute spend.
OpenAI vs Anthropic: Claude Pricing Competition Heats Up
Market pressure is intensifying as OpenAI and Anthropic fight for enterprise share on price as well as capability. OpenAI is reportedly preparing steep OpenAI price cuts on tokens to win customers back from Claude, a sign that usage-based pricing is now the main competitive lever for large models. According to the Wall Street Journal, the company is weighing reductions before Anthropic can make a similar move, reflecting concern over enterprise AI budgets that are swelling under heavy coding and agent workloads. Anthropic, meanwhile, has pushed its own pricing higher at the top end: Claude Fable 5 lists at USD 10 (approx. RM46) per million input tokens and USD 50 (approx. RM230) per million output tokens, twice the rate of Opus 4.8, even as developers receive temporary free access. These “Claude pricing competition” dynamics are drawing clear lines between cost, capability, and loyalty.
Enterprises Learn the Hard Way: Tokenmaxxing and Budget Shock
For many software and ecommerce firms, the new reality is that AI token costs resemble cloud bills more than SaaS subscriptions. GitHub notes that agentic workflows can plan, code, test, and retry for hours, burning tokens far beyond a human’s typing ceiling. Reports of Copilot users projecting hundreds or thousands in overages have forced teams to set strict budgets or risk productivity cliffs when tools shut off. In one widely discussed case, Uber reportedly blew through its entire AI budget for 2026 in four months after encouraging heavier AI use, while some staff embraced “tokenmaxxing” with long prompts and nested agent loops to climb internal usage leaderboards. This environment is pushing finance teams to monitor dashboards daily and driving conversations about whether current AI spend delivers enough value to justify the unpredictable, usage-based pricing model now standard across enterprise AI stacks.
DeepSeek and the Rise of Cheaper Alternatives
As token bills swell, enterprises are testing cheaper models to keep usage-based pricing under control, even if that means moving away from marquee providers. Axios reporting, summarized by Wccftech, indicates Microsoft is weighing a self-hosted deployment of DeepSeek’s V4 model for Copilot Cowork, which currently leans on OpenAI and Anthropic. The driver is cost: soaring token usage, especially for coding tasks and agentic workflows, is becoming a critical inhibitor for enterprise projects. Some customers complain that OpenAI and Anthropic are not only increasing enterprise pricing but also adding creative token caps, which undercuts large-scale automation plans. In response, companies are exploring open-source or lower-cost models like DeepSeek to keep AI token costs predictable while still supporting complex workloads. This shift shows how quickly loyalty can bend when the monthly bill becomes the main risk to enterprise AI budgets.
Maturing Market: Cost Efficiency Rivals Capability
Token wars are a sign that generative AI is entering a more mature, cost-sensitive phase. The early rush to add AI coding tools, chatbots, and agents has given way to a harder look at token usage, price tiers, and budget controls. GitHub’s preview bill experience and user-level spending caps, along with usage caps reported at firms like Meta, Uber, and Salesforce, show that “tokenomics” is now an operational discipline. Some, like 8x8, report they remain in the black while using Anthropic’s Claude to write emails, analyze customer feedback, and generate code, but they are the exception that proves the rule: AI must earn its keep. More aggressive OpenAI price cuts or Claude pricing competition will likely continue, yet both providers already spend heavily on infrastructure, so margins are under pressure. Future winners will balance capability with clear, predictable AI token costs.
