Codex Removes 5-Hour Limit, Fable 5 Subscription Extended Again as AI Compute Competition Heats Up

On July 9, OpenAI announced the temporary removal of the 5-hour usage limit for all paid Codex users, along with efficiency optimizations for GPT-5.6 Sol and a usage reset. Just one hour later, Anthropic extended the Claude Fable 5 subscription deadline to July 19 and increased Claude Code's weekly rate limit by 50%. These moves come in response to massive user backlash, with many users posting high bills and threatening to switch to competitors.

Background

• OpenAI: Removed the 5-hour limit for Codex Plus, Business, and Pro plans (duration unknown); introduced GPT-5.6 Sol efficiency optimizations to reduce consumption; announced 6 million active users and reset usage.
• Anthropic: Extended Fable 5 subscription to July 19 (already delayed once); increased Claude Code weekly limit by 50%; other models (e.g., Opus 4.8) also saw a 25% increase.

Key Details

• Users posted screenshots of high credit bills and cancellation confirmations, explicitly stating they would switch to OpenAI's GPT-5.6 Sol, xAI's Grok 4.5, or Meta's models.
• One developer was hospitalized after working continuously due to unused tokens.
• OpenAI CEO Sam Altman added to the prize pool but did not disclose the amount.

Analysis of Compute Supply-Demand Imbalance

• Tech analyst Benedict Evans noted that current AI inference business has 40%-50% gross margins, but this does not account for the massive spending on training next-generation models (far exceeding revenue).
• Over $1 trillion in data center capital expenditure is under construction, but new models' compute demand is growing faster, making the supply-demand crossover point uncertain.
• A specific cause of compute shortage in H1 2026 is the sudden product-market fit for software development use cases. If consumer-grade scenarios with hundreds of millions of daily active users emerge, existing infrastructure cannot support them.
• In terms of competition, frontier models are converging in methods, data, and capabilities, with no network effects or winner-take-all barriers yet. Evans predicts that once supply eases, foundation models will trend toward low-margin commoditization.

Reactions and Impact

• Developers generally welcome the changes, but some users believe OpenAI's removal of the 5-hour limit is limited in effect, with consumption optimization and reset being more practical; Anthropic's 50% weekly limit increase is a real benefit for heavy users.
• The business battle is fierce: OpenAI and Anthropic are competing for users in the most direct way—one extends deadlines, the other removes limits; one offers 50% more quota, the other removes time caps entirely.
• The tighter the compute, the more aggressive the spending, as neither side wants to lose users by easing up.