OpenAI to launch GPT-5.6 after making ‘many changes’

WASHINGTON - ChatGPT-maker OpenAI was set to release its most powerful artificial intelligence model to the public on July 9, with its chief executive saying the company made “many changes” to the system during a security review with the US government.

Chief executive officer Sam Altman told CNBC that US officials had tested the model and flagged concerns during the process.

“It was a collaborative back and forth, but we made many changes through the process,” Altman said.

The company’s new GPT-5.6 offerings and other cutting-edge AI models, including Anthropic’s Mythos series, have drawn concern over what researchers have described as an unprecedented ability to identify weaknesses in code that hackers can exploit.

That has raised national security fears, and OpenAI said in late June it had shared preview access to GPT-5.6 with a limited group of trusted US-based partners at Washington’s request.

The GPT-5.6 series has three tiers: Sol, the company’s new flagship model; Terra, a mid-range version for everyday work; and Luna, a fast, low-cost option.

Altman on July 9 touted Sol’s performance, saying it is “not only the best model in the world for most people” but also “54 per cent more token efficient on tasks” than competing models.

Tokens are the industry unit of AI usage and are often used for pricing for customers.

The Trump administration earlier this week gave OpenAI the green light for the broad launch of GPT-5.6, following technical testing and meetings between the company and government officials.

The White House, wary of seeming pro-regulation, officially denies that the process amounts to an approval, saying AI companies submit their models for scrutiny on a voluntary basis.

The delay in release echoes a similar situation at OpenAI’s archrival Anthropic.

The startup behind the Claude chatbot, which has a tense relationship with the Trump administration, last week began restoring access to its most powerful AI models after Washington lifted a restriction on releasing them to foreigners.

Meanwhile, Instagram and Facebook giant Meta this week released its major AI models, some of the first since CEO Mark Zuckerberg overhauled the company’s artificial intelligence team.

Zuckerberg on July 8 announced the launch of Muse Spark 1.1 on X, describing it as “a strong agentic and coding model at a very low price.”

The release signals Meta’s push into the market for AI models for computer coders, putting it in more direct competition with OpenAI and Anthropic as the commercial stakes in the AI race continue to climb.

Meta on July 7 released Muse Image, an AI image generator that has already drawn controversy as it allows users to manipulate Instagram users’ images, as long as that profile is public. AFP