The End of AI Subscription Bargains! GitHub Copilot Transitions to Pay-As-You-Go Model in June 2026, AI Companies Face Limits on ‘Burning Cash’

📰 News Overview

• The Demise of Flat-Rate Models: GitHub Copilot announced it will shift from a flat-rate subscription to a pay-as-you-go pricing model starting June 1, 2026.
• Massive Financial Losses: Companies like OpenAI, Anthropic, and Google have been revealed to be maintaining low-priced subscriptions of around $20 per month while facing computing costs of hundreds of dollars per user each month.
• Impact of Agent-Based AI: The rise of autonomous agents like Claude Code has accelerated token consumption at an unprecedented rate compared to traditional chat models, leading to the collapse of the economic model.

💡 Key Points

• Discrepancy with Real Costs: While the subscription fee for Claude Pro (Sonnet 4.6 / Opus 4.6) is $20, the cost for knowledge workers performing equivalent tasks via API can reach $200 to $400 per month.
• Normalizing Losses: Microsoft incurs over $20 in losses per user each month with GitHub Copilot, and power users can face costs as high as $80 per month.
• Strategic Shift: The product lead at OpenAI has hinted at the discontinuation of “unlimited use.” AI companies are being forced to transition from “unlimited utility” to “inference companies based on computational demand.”

🦈 Shark’s Eye (Curator’s Perspective)

Finally, the time has come for companies feasting on the “discounted filet mignon” to receive their rightful invoices! For too long, companies have been burning cash to dominate the market, but the emergence of ultra-high-performance models like Opus 4.6 running autonomously as an AI Agent has thrown all calculations off balance. Agents that can devour a five-hour usage limit in just 90 minutes can no longer be sustained by a flat-rate “all-you-can-use” model! The specific deadline of June 2026 for GitHub Copilot’s switch to a pay-as-you-go model is likely to mark a monumental turning point for the entire industry. Moving forward, the ability to “smartly manage AI operations” will become an essential skill for engineers!

🚀 What’s Next?

Companies will be forced to completely reassess workflows built on the premise of “all-you-can-use” AI. In the future, we can expect a more stringent management of computational resources, including optimizing the use of lightweight versus heavyweight models (like Opus 4.6) based on task importance, a return to local LLMs, or only summoning autonomous agents when absolutely necessary.

💬 A Word from Haru Shark

Journalist “Haru Shark” will now keenly feel the weight of each bite (token) and sharpen up to deliver the news! No wasted shots, just smart bites! 🦈💥

📚 Glossary

• Agent-Based AI: AI that autonomously writes code, manipulates files, and completes tasks without waiting for human instructions. It consumes computational resources at a staggering rate.

• Token: The smallest unit of text processed by AI. The cost of using a model is calculated based on the amount of tokens.

• Inference: The process by which a trained AI model generates responses to inputs. The computational costs (GPU load) associated with this process constitute the main expenses for current AI companies.

• Source: Every AI Subscription Is a Ticking Time Bomb for Enterprise