OpenAI Releases GPT-5.1-Codex-Max That Independently Performs Coding Tasks

OpenAI has unveiled GPT-5.1-Codex-Max, a frontier agentic coding model designed to autonomously handle complex software engineering tasks across multiple stages of the development lifecycle.

Built on an updated foundational reasoning model, this iteration represents a significant advancement in AI-assisted coding capabilities, combining improved intelligence with enhanced token efficiency and multi-context window processing.

Advanced Long-Context Processing and Autonomous Capabilities

GPT-5.1-Codex-Max introduces native multi-context window processing through a process called compaction, enabling the model to coherently work across millions of tokens in a single task.

This breakthrough unlocks previously impossible workflows including project-scale refactors, extended debugging sessions, and multi-hour autonomous agent loops.

The model can sustain work for extended periods internal evaluations show GPT-5.1-Codex-Max operating independently for more than 24 hours while persistently iterating on implementations, fixing test failures, and delivering successful results.

The model has been trained specifically on real-world software engineering tasks including pull request creation, code review, frontend development, and quality assurance operations.

Notably, GPT-5.1-Codex-Max is the first model trained to operate natively in Windows environments, expanding its practical deployment scenarios.

Performance improvements are substantial: on SWE-bench evaluations, the model achieves 79.9% accuracy compared to GPT-5.1-Codex’s 66.3%.

Token efficiency improvements translate directly to cost savings for developers. On SWE-bench Verified tasks, GPT-5.1-Codex-Max with medium reasoning effort achieves superior performance while consuming 30% fewer thinking tokens than its predecessor.

The model introduces an Extra High reasoning effort setting for non-latency-sensitive workloads, enabling deeper analysis when extended processing time is acceptable.

Frontend design generation exemplifies these efficiency gains: GPT-5.1-Codex-Max produces functionally equivalent interfaces using only 27,000 thinking tokens compared to 37,000 for GPT-5.1-Codex, with comparable aesthetic quality and fewer tool calls.

OpenAI acknowledges the dual-use nature of advanced coding capabilities. While GPT-5.1-Codex-Max does not reach High capability on cybersecurity assessments under the Preparedness Framework, it represents the most advanced cybersecurity model deployed to date.

The company has implemented dedicated cybersecurity monitoring to detect malicious activity and has already disrupted cyber operations attempting to misuse the model.

Codex operates in a secure sandbox by default with limited file access and disabled network functionality.

OpenAI recommends maintaining this restricted mode, as enabling internet access introduces prompt-injection risks from untrusted content.

The model generates terminal logs and cites tool calls, supporting human review before production deployment.

GPT-5.1-Codex-Max is available through Codex CLI, IDE extensions, cloud platforms, and code review tools, with API access coming soon.

The model replaces GPT-5.1-Codex as the default option across Codex surfaces for ChatGPT Plus, Pro, Business, Edu, and Enterprise plans.

Product Details Table

Aspect	Details	Risk Level
Model Name	GPT-5.1-Codex-Max	Low
Deployment Method	Codex CLI, IDE, Cloud, API	Low
Context Window Processing	Multi-context via compaction; 24+ hour autonomy	Medium
Cybersecurity Capability	Not High; most advanced deployed; monitored	Medium-High
Sandbox Security	Default restricted; file/network limited	Low
Human Review Requirement	Mandatory before production deployment	Operational
Threat Mitigation	Prompt-injection risk; untrusted content exposure	Medium
Attack Surface	Network access, external integrations, code execution	High

Find this Story Interesting! Follow us on Google News, LinkedIn and X to Get More Instant Updates

The post OpenAI Releases GPT-5.1-Codex-Max That Independently Performs Coding Tasks appeared first on Cyber Security News.