OpenAI Rolling Out GPT-5.2 Codex-Max to Select Users

OpenAI has begun distributing what appears to be the next evolution of its agentic coding model to a limited number of paid subscribers. Users have discovered the rollout by directly querying Codex about which model powers their session - with some instances now identifying themselves as GPT-5.2-Codex-Max rather than the standard GPT-5.2-Codex released in December 2025.

The appearance of a Max variant follows the pattern OpenAI established with GPT-5.1, where the Codex-Max version delivered substantial performance improvements over the base model. If history repeats, GPT-5.2-Codex-Max could represent another significant leap in agentic coding capabilities.

The Codex Model Hierarchy

Understanding where Codex-Max fits requires context on the current GPT-5.2 Codex lineup. The base GPT-5.2-Codex model is available to all paid ChatGPT users and represents the standard agentic coding experience. GPT-5.2-Codex-Max appears positioned as the enhanced variant with additional capabilities, currently in limited rollout for testing.

The Max designation has historically indicated higher reasoning effort limits, better performance on complex multi-step tasks, enhanced context handling for large repositories, and improved accuracy on professional benchmarks. These improvements compound significantly for professional software engineering workflows.

What GPT-5.2-Codex Already Delivers

The base GPT-5.2-Codex model - what OpenAI describes as their most advanced agentic coding model yet - already represents a substantial capability jump. The model excels at long-horizon work through intelligent context compaction, maintaining coherence across extended coding sessions without losing track of overall project structure or previous decisions.

Large-scale code changes that previous models struggled with have become routine. Refactoring entire codebases without breaking functionality, migrating between frameworks or API versions, and implementing complex features spanning multiple files all fall within the model's operational capabilities.

Windows environment support addresses a gap that frustrated developers working outside Unix-like systems. The model now performs reliably in native Windows development environments. Enhanced vision capabilities allow interpretation of screenshots showing UI bugs, technical diagrams, design mockups, and chart analysis during coding sessions.

Benchmark performance tells the technical story. GPT-5.2-Codex achieves state-of-the-art results on SWE-Bench Pro for real-world software engineering tasks and Terminal-Bench 2.0 for agentic terminal performance. The model scores 87% accuracy on CVE-Bench for vulnerability research and achieves the highest scores in the GPT family on capture-the-flag security challenges.

The Cybersecurity Dimension

One of the most significant aspects of GPT-5.2-Codex is its substantially improved cybersecurity capabilities. OpenAI explicitly calls it their strongest cybersecurity model yet - a characterization with real-world validation.

Security researcher Andrew MacPherson at Privy used the previous GPT-5.1-Codex-Max to discover and responsibly disclose a source code exposure vulnerability in React, a framework underpinning millions of applications. The capabilities extend to multi-step security analysis in capture-the-flag style challenges, fuzzing and test environment setup, attack surface analysis, malware analysis for remediation, and infrastructure stress testing.

To mitigate risks, OpenAI is rolling out gradually with safeguards, piloting invite-only trusted access for vetted security professionals, working with red teaming organizations, and implementing specialized safety training against harmful tasks and prompt injections.

What Codex-Max Likely Adds

Based on the pattern established with GPT-5.1-Codex-Max, the new Max variant likely includes extended reasoning effort capabilities. The previous Max introduced an Extra High reasoning effort level allowing the model to think longer on complex problems. GPT-5.2-Codex-Max may push this further or make extended reasoning the default.

Max variants have historically demonstrated improved token efficiency, delivering better answers while consuming fewer resources - important for long coding sessions where context accumulates. Better reliability on edge cases means the Max variant typically shows improved consistency and fewer failures where the base model might falter on unusual or highly complex tasks.

Enhanced agentic behavior represents perhaps the most significant improvement. For multi-step tasks requiring the model to plan, execute, observe results, and iterate, Max variants demonstrate more robust agentic loops with better error recovery and adaptation.

Checking Your Access

Paid ChatGPT subscribers can check for Codex-Max access through several methods. In the Codex CLI, starting a session and using the /model command or asking "What model are you currently using?" will reveal the active model. IDE extension users can check the model dropdown in their extension settings where GPT-5.2-Codex-Max will appear as an option if access has been granted.

For those wanting to explicitly request the Max model when available, updating the configuration file to specify gpt-5.2-codex-max as the model will route requests appropriately once general availability begins.

Enterprise Momentum

OpenAI reports that Codex usage has grown more than 20x since the launch of GPT-5-Codex in August. Major enterprises now using Codex include Cisco, Ramp, Virgin Atlantic, Vanta, Duolingo, and Gap. For enterprise engineering teams, the progression to GPT-5.2-Codex-Max represents continued improvement in AI-assisted development workflows.

The rapid iteration cycle - GPT-5.0 to 5.1 to 5.2 in months - suggests continued rapid improvement. Teams evaluating AI coding assistants should recognize that GPT-5.2-Codex-Max appears positioned as the premium tier for professional software engineering, while cybersecurity teams should watch the trusted access pilot for defensive research capabilities.

Timeline Expectations

The limited rollout is happening now in January 2026. An official announcement is expected in the coming days, with wider availability to paid users likely within one to two weeks. API access is expected in the coming weeks according to OpenAI's typical release patterns.

Strategic Implications

For developers, early access matters because users seeing Codex-Max are getting preview access to capabilities that will eventually reach everyone. This limited rollout allows OpenAI to gather real-world feedback before wider release while maintaining competitive pressure against Anthropic's Claude, Google's Gemini, and other platforms pushing hard on coding capabilities.

The trajectory suggests AI-assisted development is becoming increasingly capable and increasingly essential to remain competitive. For enterprises already using Codex, GPT-5.2-Codex-Max validates that the platform continues to advance rapidly. For those evaluating options, the progression demonstrates that standing still means falling behind.