OpenAI’s Leaked GPT‑5.5, New Agent Platforms, and Claude Code Controversy

The AI race just picked up again. OpenAI appears to have accidentally leaked several new models, including GPT‑5.5, while also rolling out an enterprise agent platform. Google is pushing hard with its own Gemini Enterprise agent stack, and Anthropic has stirred controversy by quietly testing the removal of Claude Code from its Pro plan. On top of that, Google says 75% of its new code is now written by AI.

Catch all the details in the video below, then dive into the breakdown.

OpenAI Leak: GPT‑5.5 and New “Glacier Alpha” Models

OpenAI briefly exposed a set of internal model names, revealing what looks like the next wave of its frontier systems. The standout entry is GPT‑5.5, described as “the latest frontier agentic coding model.” That suggests a strong focus on autonomous coding and agent-style behavior, not just better chat.

The leak also referenced the current GPT‑5.4 and a model called OAI 2.1, which carries the same “frontier agentic coding” description. OAI 2.1 might be a lighter or more specialized variant, but for now, it’s just a name with a hint of its purpose.

Alongside those, several codenames appeared:

Arcane – labeled a “frontier model with legendary appetite for starches” (clearly a playful internal description).

Glacier Alpha – called “the intelligence that moves continents,” which had previously been rumored as the codename for GPT‑5.5, but now seems to be tracked separately.

Glacier Alpha Block CUI 3 and Glacier Alpha Block CUI 4 – both described as “ice cold intelligence.” These could be internal checkpoints, experimental variants, or future-tier models (similar in spirit to how Anthropic has Haiku, Sonnet, and Opus).

Until OpenAI makes an official announcement, all of this remains speculative. However, the presence of GPT‑5.5 in internal systems strongly suggests a near‑term release, especially given OpenAI’s recent push toward agentic workflows and coding automation. For a broader view of how these frontier labs are evolving, you may also want to check out this roundup of major updates from Anthropic, OpenAI, Google, and others.

OpenAI Workspace Agents: An Agent Studio for Enterprises

OpenAI has officially launched Workspace Agents in ChatGPT, an agent-building environment aimed squarely at enterprise customers. The idea is simple: let companies spin up AI agents that automate repetitive workflows, then share and manage them across teams.

From day one, OpenAI is shipping a set of prebuilt agents, including:

Tally – generates weekly reports.

Spark – qualifies leads, sends follow‑ups, and supports CRM workflows.

Scout – turns user feedback into prioritized tickets and insights.

Enterprises can also create their own agents by describing workflows and desired behaviors. These agents are powered by OpenAI’s Codex stack, so they can handle complex logic and integrations from the start.

Key capabilities include:

Easy sharing across teams – a successful agent built by one team can be shared with others inside the same organization.

Integration with tools like Slack – for example, a Scout agent can sit in a Slack channel and answer questions like “What’s the latest feedback on our mobile app?”

Governance and compliance controls – essential for enterprises that need visibility into what agents are doing, plus guardrails around data and actions.

OpenAI also teased what’s coming next: new triggers to start work automatically, improved dashboards to track performance, more integrations with business tools, and support for workspace agents inside the Codex desktop app. If you’re interested in building with Codex specifically, the deep‑dive course covered in this guide to the Codex desktop agent is a useful complement.

For now, Workspace Agents are focused on enterprise customers, not general consumers. That aligns with OpenAI’s broader strategy: lock in large organizations with powerful, integrated AI workflows that are hard to replace once adopted.

Gemini Enterprise: Google’s End‑to‑End Agent Platform

OpenAI isn’t alone in chasing the enterprise agent market. Google is rolling out a major expansion of Gemini Enterprise, positioning it as an end‑to‑end platform for the “agentic era.” It’s effectively the next evolution of Vertex AI, but with a stronger focus on agents and collaboration.

New capabilities for Gemini Enterprise users include:

Project Skills and Agent Builder – tools to design, configure, and deploy custom AI agents.

Agents Gallery – a catalog of ready‑made agents (for tasks like synthesis, fact‑checking, deep research, and idea generation) that teams can quickly adopt instead of building from scratch.

Slides Editor inside Canvas – AI‑assisted slide creation and editing directly within Google’s collaborative environment.

Gemini Enterprise is designed so teams can discover, create, share, and run agents in one secure environment. Google is also building an open partner ecosystem so enterprises can deploy third‑party agents from major vendors like Oracle, Salesforce, and ServiceNow.

The pattern is clear: frontier labs are racing to become the default AI operating layer for large organizations. Enterprise customers bring recurring revenue, long‑term contracts, and deep integration, all of which matter a lot as companies like OpenAI and Anthropic move toward potential IPOs.

Anthropic’s Claude Code Backlash

While OpenAI and Google are announcing new platforms, Anthropic has found itself in a mini‑crisis over Claude Code, its coding and agent environment. Users noticed that Claude Code appeared to be quietly removed from the Claude Pro plan for some new sign‑ups, without a clear public announcement.

According to Anthropic’s head of growth, this change was part of a small test affecting about 2% of new subscribers. Existing Pro and Max users were not supposed to be impacted. The explanation was that when the Max plan launched, it was intended for heavy chat use only, and features like Claude Code, Code Work, and long‑running agents were added later, changing how people used the subscription.

Even with that clarification, many users were frustrated by the lack of transparency. The move also drew competitive fire: OpenAI’s Codex team publicly emphasized that Codex would remain available on both free and Plus ($20) plans, saying they have the compute and efficient models to support it and stressing that “transparency and trust” are principles they won’t break, even at the cost of short‑term revenue.

Underlying all of this is a practical issue: compute capacity. Claude users frequently hit rate limits, which suggests Anthropic is still constrained on infrastructure. Removing or limiting high‑compute features like Claude Code for some users may be a way to manage those constraints, but doing it quietly has clearly hurt trust. For more context on Anthropic’s recent direction and concerns around its models, you can read this analysis of Anthropic’s Mythos model.

Google Says 75% of New Code Is Now AI‑Generated

In a blog post tied to Google Cloud Next ’26, CEO Sundar Pichai shared a striking statistic: 75% of all new code at Google is now generated by AI. Engineers are increasingly acting as orchestrators of autonomous AI agents rather than writing every line of code by hand.

This doesn’t mean human developers are obsolete, but it does show how fast the industry is changing. The products billions of people use every day are now largely built with AI‑assisted or AI‑generated code. For anyone following AI closely, it’s a powerful signal of where software development is heading: humans designing systems and guardrails, AI handling most of the implementation details.

Put together with the OpenAI leaks, enterprise agent platforms from both OpenAI and Google, and Anthropic’s struggles to balance features with compute limits, the message is clear: we’re moving into an era where agents and AI‑generated code are the default, not the exception.