The Vault -- AI Edition | May 3, 2026 | 4 min read | 872 words

Category: business

Key Takeaway: OpenAI's GPT-5.5 release and the rapid ascent of Codex to four million weekly users marks a pivotal shift from chatbot novelty to agentic work platform, with pricing that challenges developers to rethink their AI budgets.

---

OpenAI shipped GPT-5.5 on April 23 and within ten days the numbers told a story the company has been chasing since ChatGPT first went viral: this time, the users are staying and they are paying. Codex, the GPT-5.5-powered agentic coding and work automation tool, has crossed four million weekly active developers, up from three million just weeks earlier. CEO Sam Altman described the model as the last major milestone before artificial general intelligence, a claim that would sound grandiose from anyone else but lands differently when backed by the fastest-growing product launch in OpenAI's history.

The Model: Bigger, Pricier, and Built for Agents

GPT-5.5, codenamed Spud internally, is not merely an incremental step over its predecessor. Running on NVIDIA GB200 NVL72 rack-scale systems, the model ships with a 400,000-token context window and is available across Plus, Pro, Business, Enterprise, Edu, and Go plan tiers within ChatGPT. API access followed one day later on April 24.

The pricing tells its own story. OpenAI doubled the per-token cost on the GPT-5 line: input tokens now run $5 per million, up from $2.50 on GPT-5.4, while output tokens jumped to $30 per million from $15. For developers watching their margins, a batch processing endpoint at 50 percent of standard pricing offers some relief, and a new cached input rate of $0.50 per million tokens targets repetitive workloads. A premium tier, GPT-5.5 Pro, commands $30 per million input tokens and $180 per million output tokens for high-accuracy use cases.

The price hike is a calculated bet. OpenAI is signaling that GPT-5.5 delivers enough marginal value in code generation, scientific reasoning, and multi-step planning to justify the markup. Early benchmark results appear to support that confidence, with the model outperforming GPT-5.4 across coding, mathematics, and general knowledge evaluations.

Codex: From Coding Agent to Super App

The more consequential story may be what GPT-5.5 powers rather than what it scores on benchmarks. Codex, which OpenAI first launched as a software engineering agent in May 2025, has evolved into something far more ambitious. The tool now operates as a full-fledged work agent capable of browsing the web, generating images, scheduling tasks, and controlling desktop applications autonomously on macOS through background computer use.

"Codex is becoming the operating system for knowledge work," Altman wrote in an internal memo later shared publicly by NVIDIA, whose roughly 10,000 employees received early enterprise access to the GPT-5.5-powered version spanning engineering, legal, marketing, finance, and human resources departments.

With over 90 integrated plugins ranging from Atlassian to the Microsoft Suite, Codex has outgrown the coding-assistant label that defined its early days. OpenAI is positioning it squarely as a super app, a term the company has increasingly embraced in public communications.

The growth trajectory is striking. Codex hit three million weekly users in early April, prompting Altman to reset usage limits in celebration and pledge to repeat that gesture at every additional million users until reaching ten million. Reaching four million barely two weeks later suggests the GPT-5.5 upgrade is accelerating adoption rather than merely retaining existing users.

The Competitive Landscape Tightens

OpenAI's aggressive push comes as the competitive field intensifies. Anthropic's Claude Code has been gaining traction in developer workflows, while Google's Gemini models continue to close the gap on reasoning benchmarks. The decision to double API pricing is a bold move in this environment, essentially daring competitors to match capability rather than undercut on cost.

"OpenAI is making a very clear bet that developers will pay more for a model that can actually do the work, not just talk about it," said Swyx Shawn Wang, founder of the AI Engineering community and a close observer of the developer tools market. "The shift from chatbot to agent changes the value equation entirely."

Enterprise adoption signals reinforce this thesis. NVIDIA's company-wide rollout is not a pilot program but a production deployment, and the speed at which it was executed suggests the GPT-5.5 upgrade was coordinated well in advance of the public launch.

What to Watch

Three questions will determine whether this launch sustains momentum. First, whether the doubled API pricing holds or whether competitive pressure forces concessions before the end of Q2. Second, whether Codex's expansion beyond coding into general work automation creates genuine retention or merely inflates weekly active user counts with casual experimentation. Third, whether Altman's framing of GPT-5.5 as the last stop before AGI sets expectations the next release cannot meet.

For now, four million weekly users and a model that enterprises are deploying at scale give OpenAI its strongest hand in months. The question is no longer whether AI agents will reshape knowledge work but which platform will own that transformation.

---

Sources:

"OpenAI is making a very clear bet that developers will pay more for a model that can actually do the work, not just talk about it."
-- Swyx Shawn Wang, Founder, AI Engineering community
4M
Codex weekly active developers
$5/$30
API pricing per million tokens
400K
Context window tokens