After the launch of GPT-5.6, the combined weekly active users of Codex and ChatGPT Work exceeded 8 million, with user base growing more than sevenfold in five months. The explosive growth strained OpenAI’s infrastructure, forcing it to reduce context windows and optimize inference capacity. OpenAI is simultaneously advancing its platform strategy by integrating Codex into the ChatGPT desktop app, but usage quota constraints remain a core challenge.Article author and source: Wall Street Journal
OpenAI’s AI coding tool Codex is experiencing a rare surge in user growth. Since the launch of GPT-5.6, the combined weekly active users of Codex and ChatGPT Work have surpassed 8 million within days, with weekly usage increasing 2.5 times compared to previous levels. Over the past five months, the total user base has expanded more than sevenfold, outpacing growth rates across the entire enterprise SaaS industry.
Later, he posted again, calling the growth of GPT-5.6 Sol "incredible," praising the reasoning team's efforts to meet demand, and acknowledging that some service fluctuations may occur in the near term.
The surge in users directly strained OpenAI’s infrastructure and forced competitor Anthropic to respond swiftly—within hours of OpenAI announcing its milestone of 7 million users, Anthropic extended its promotional pricing for Claude Fable 5 and increased the weekly usage limit for Claude Code by 50%. The competitive landscape in AI-powered programming is accelerating its transformation.
GPT-5.6 ignites the growth engine; Codex's user growth curve is nearly vertical.
According to Tibo Sottiaux, Engineering Lead at OpenAI Codex, Codex had fewer than 1 million weekly active users in February, rising to 5 million by early June. After the official launch of GPT-5.6 on July 9, growth accelerated sharply: reaching 6 million on July 12, surpassing 7 million approximately 24 hours later, and hitting 8 million by Sunday.
Sottiaux announced this milestone on social platform X, simultaneously resetting usage limits for all users while maintaining the previously lifted 5-hour usage cap policy, encouraging users to explore the capabilities of GPT-5.6 Sol.
The direct catalyst for this growth was OpenAI’s integration of multiple products on the same day: incorporating the standalone Codex app into the ChatGPT desktop application, launching a new agent mode called ChatGPT Work tailored for knowledge workers, and beginning the phased sunsetting of the Atlas browser. This consolidation of the product lineup significantly lowered the barrier to entry for users.
The traffic surge exposed the limits of scaling. The influx of users almost immediately strained OpenAI’s infrastructure. According to the company, traffic within 48 hours of launch reached approximately twice the previous peak, revealing several scaling issues.
On July 12, Sottiaux released a detailed explanation outlining the team's response measures: increasing available capacity per subscriber by approximately 10% through optimized inference; reducing the context window from 372,000 tokens to 272,000 tokens (the previous higher limit had caused unexpected billing issues); rolling back the experimental inference intensity setting internally codenamed "juice"; and fixing overly aggressive multi-agent behavior at the highest inference level.
Additionally, OpenAI has temporarily removed the five-hour usage limit for Plus, Business, and Pro subscribers—the most open access policy since the product's launch.
The above adjustments have sparked division within the community: some users interpreted the reduction in context window as a de facto downgrade, while others praised Sottiaux for openly explaining the operational trade-offs.
Sam Altman then stepped in, framing OpenAI's approach as a stance against competitors' dismissive treatment of users.
The platform strategy is becoming a core competitive advantage. Codex’s explosive growth reflects OpenAI’s deeper shift in product strategy. By integrating Codex, ChatGPT Work, and a built-in browser into a single desktop application, and enabling connectivity through a plugin architecture to mainstream tools such as Slack, Google Drive, SharePoint, CRM, and calendars, OpenAI is transforming the ChatGPT desktop experience into a unified workspace layered atop existing workflows.
This logic is also reflected in the dimension of cost efficiency. For example, Cognition’s Devin Fusion, powered by Fable 5, achieves a lower cost per task than Anthropic’s more expensive Opus 4.8, due to more efficient task allocation that reduces wasted effort: in 81% of runs led by Fable, the primary model never directly edited code. This case demonstrates that a higher-priced model may actually reduce overall costs if it effectively avoids resource waste.
Usage limits remain a bottleneck despite strong growth. Although the five-hour limit has been removed, Codex and ChatGPT Work still share the same weekly usage pool. When users run intensive Sol Ultra multi-agent orchestration tasks, their quota is consumed rapidly. Sottiaux has publicly acknowledged this issue and stated that a fix is currently in progress.
While Sam Altman confirmed the growth data, he also acknowledged that recent service fluctuations may occur as demand continues to rise. Analysis suggests that for developers and enterprise users betting on AI programming tools, infrastructure scalability may be as critical as product capability in the second half of this growth race.
It is noteworthy that the rapid increase in user data has directly impacted the competitive landscape.
Within hours of OpenAI announcing a breakthrough of 7 million users, Anthropic extended the promotional pricing for Claude 3.5 to July 19 and increased the weekly usage limit for Claude Code by 50%. The timing of these two announcements closely aligns, making the market’s interpretation of their connection self-evident.
On performance benchmarks, after completing 7,800 real-world agent tasks, GPT-5.6 Sol has risen to second place on the Arena agent leaderboard, prompting many developers to once again view OpenAI as the leader in AI programming.
On Anthropic’s side, the launch of the Claude Cowork product also stems from users spontaneously using Claude Code for broader applications beyond software development, indicating that both companies are evolving their product formats in similar directions.
From a broader competitive perspective, OpenAI holds a scale advantage with its massive base of 900 million weekly active users and leading benchmark scores; Anthropic maintains deep traction among developers through its reputation for high-quality code and enterprise-friendly pricing; for users prioritizing data sovereignty, open-source solutions running GLM 5.2 or Kimi K2.7 offer a more attractive alternative, albeit at the cost of some speed.