(CTN News) – On Friday, OpenAI declared that Codex, their most advanced AI coding agent to date, will be accessible for research purposes.

Codex operates on Codex-1, a modified version of the company’s O3 AI reasoning engine specifically designed for software engineering. OpenAI asserts that Codex-1 excels at adhering to instructions, generates more refined code than O3, and conducts iterative testing of its code until it achieves success.

Cloud-based virtual machines are sandboxed to ensure the secure operation of the Codex agent. Linking to GitHub enables the preloading of your code repositories into the Codex environment.

OpenAI states that the AI coding agent requires between one to thirty minutes to perform tasks such as answering enquiries about your codebase, executing tests, rectifying bugs, and developing fundamental features.

Aside from providing browser support, Codex is also computer-based.

Commencing today, subscribers to ChatGPT Pro, Enterprise, and Team will gain access to Codex. OpenAI states that users will initially receive “generous access” to Codex, although rate limits will be implemented subsequently.

An OpenAI representative has notified TechCrunch that users will have the option of purchasing additional credits to continue utilising Codex.

Codex will soon be available to Edu and ChatGPT Plus users.

The term “vibe coders” has recently gained popularity as an alternative designation for artificial intelligence tools designed for software engineers.

According to the CEOs of Google and Microsoft, artificial intelligence currently occupies approximately 30% of their code. In February, Anthropic launched their agentic coding tool, Claude Code, while Google enhanced their AI coding assistant, Gemini Code Assist, in April by incorporating new agentic features.

The popularity of AI coding platforms has led to rapid growth rates for the companies that operate them in the IT sector. Allegedly pursuing additional funding at a $9 billion valuation, the prominent artificial intelligence coding tool Cursor generated approximately $300 million in April.

Vigorous competition for market share is currently underway at OpenAI. Rumours suggest that the company responsible for ChatGPT has consented to pay $3 billion to acquire Windsurf, the creators of another prominent AI coding platform.

Furthermore, OpenAI is evidently advancing its proprietary AI coding instruments with the launch of Codex. Codex users can assign the agent new coding tasks by inputting a prompt and selecting the “Code” button.

The tool is located in the sidebar of ChatGPT. A user may submit enquiries regarding their codebase by selecting the “Ask” button. Users can track Codex’s advancement and view additional assignments through the prompting bar.

In a briefing prior to Codex’s launch, Josh Tobin, OpenAI’s Agents Research Lead, informed TechCrunch that the company’s objective is for its AI coding agents to operate as “virtual teammates”, autonomously completing tasks that would require human engineers “hours or even days” to finish.

Codex is used by OpenAI to draft documentation and scaffold features.

OpenAI’s Product Lead, Alexander Embiricos, stated that “A significant portion of the safety efforts for the company’s O3 model is relevant to Codex as well.”

An article on OpenAI’s blog asserts that the Codex review board will consistently reject proposals for the development of “malicious software.”

Moreover, Codex operates within an air-gapped environment, indicating it is not connected to the broader internet or external APIs. Administering Codex in this manner diminishes the likelihood of adverse effects, although it may concurrently reduce its efficacy.

It is important to remember that AI coding agents are prone to errors, similar to other generative AI systems. A recent study by Microsoft revealed that Claude 3.7 Sonnet and O3-mini, two prominent AI coding models, faced challenges in consistently debugging software. The enthusiasm of investors for these instruments remains unabated.

Codex CLI has recently been launched as an open-source coding agent for terminal use, now featuring a software engineering-orientated variant of OpenAI’s OP-mini model.

The standard model in Codex CLI is presently priced at $1.50 per 1,000,000 input tokens (approximately 750,000 words, exceeding the entirety of the Lord of the Rings book series) and $6 per 1,000,000 output tokens via OpenAI’s API.

Codex represents OpenAI’s most recent initiative to enhance ChatGPT’s functionalities beyond the widely recognised chatbot. OpenAI’s web browsing agent, Operator; its research agent, Profound Research; and its artificial intelligence video platform, Sora, have been accessible to subscribers with priority access for the past year.

Attracting new ChatGPT subscribers and persuading existing ones to increase their payments to OpenAI for enhanced rate limits are two potential outcomes of these services for Codex.

SOURCE: TC