Why Token Prices Are Soaring: Codex Lifts 5‑Hour Limit and Fable 5 Extends Subscription Amid Compute Crunch

The article examines recent moves by A company and OpenAI to extend Fable 5 subscriptions and remove Codex usage limits, the resulting user backlash, and Benedict Evans' analysis of a severe AI compute supply‑demand imbalance that threatens to turn foundation models into low‑margin commodities.

Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Why Token Prices Are Soaring: Codex Lifts 5‑Hour Limit and Fable 5 Extends Subscription Amid Compute Crunch

Claude Fable 5 subscription extension and Codex limit removal

Company A announced that the Claude Fable 5 subscription deadline is extended by seven days to July 19. Within an hour OpenAI removed the five‑hour usage restriction on Codex without specifying an end date.

Developers posted screenshots of high‑cost credit‑card bills and threatened to switch to OpenAI’s GPT‑5.6 Sol or xAI’s Grok 4.5, prompting Company A to grant an additional extension, yet user churn continued.

Compute supply‑demand imbalance

Benedict Evans identifies two certainties: AI inference capacity is severely undersupplied and the shortage is unsustainable.

Industry data show AI inference gross margins of 40 %–50 % after accounting for server depreciation or lease costs, but before the massive annual training expenditures that already exceed most companies’ revenues.

More than $1 trillion in data‑center capital expenditure is in the pipeline; inference efficiency is improving, yet each new model generation demands substantially more compute.

Evans notes that in the first half of 2026 software‑development workloads achieved product‑market fit, sharply increasing compute pressure. A consumer‑scale scenario with hundreds of millions of daily active users would outstrip global infrastructure at any price.

Competitive dynamics

Current frontier models share similar scientific methods, training data sources, and capabilities. Meta and xAI rebuilt their models from near‑zero within the past six months and quickly returned to top‑ranking positions, but no network‑effect or winner‑takes‑all barrier has emerged.

Analogy to the mobile‑telecom boom: massive traffic growth generated trillion‑dollar revenues while operators saw modest returns, with most value captured by upstream ecosystem players.

Long‑term implication

Unless an unforeseen shift occurs, foundation models are expected to become low‑margin, commoditized infrastructure once the current supply crunch eases.

Anthropic and OpenAI are directly competing for the same user base: Anthropic extends access, OpenAI removes limits; the first to relent may lose users.

Reference: Benedict Evans, “Ways to think about token pricing”, 2026, https://www.ben-evans.com/benedictevans/2026/7/9/ways-to-think-about-token-pricing

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

OpenAIindustry analysisAI computetoken pricingClaude Fable 5Benedict Evans
Machine Learning Algorithms & Natural Language Processing
Written by

Machine Learning Algorithms & Natural Language Processing

Focused on frontier AI technologies, empowering AI researchers' progress.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.