OpenAI quietly flexes on the agent race again

OPENAI

OpenAI quietly flexes on the agent race again

OpenAI is giving its AI agent, Operator, an upgrade.

It’ll now run on o3, one of OpenAI’s newest reasoning models, replacing the previous GPT-4o version.

The API version stays on 4o, but this update means better performance for anything that needs strong logic or maths skills.

It’s also been trained with extra safety data to help it handle tricky or sensitive requests more responsibly.

Operator remains locked behind the $200-a-month ChatGPT Pro paywall, but has expanded to most countries worldwide since its US-only launch. OpenAI describes it as a research preview.

Here’s what to know:

• Operator now uses OpenAI’s more capable o3 model.

• Safety training makes it more resistant to risky prompts.

• The AI agent space is heating up, with Google and Anthropic in the mix too.

Prompt injection? Not on my watch

Operator is part of a wider trend: AI agents that can browse the web or complete tasks on their own.

Google’s Gemini and Anthropic’s tools are doing the same.

While the new o3 Operator is better at coding, it still doesn’t get access to a full coding terminal, just like before.

Honestly, this sounds like an iPhone update, but with trust issues.