OpenAI releases GPT-5.6 Sol, its most powerful AI model to date
OpenAI has publicly released its flagship AI technology, the GPT-5.6 Sol, its most powerful model yet, along with a new tool to help with everyday office work.
The new release on Thursday escalates the battle for AI dominance with rival start-up Anthropic as the technology grows more sophisticated.
Along with its new model, OpenAI unveiled a new tool, ChatGPT Work, powered by Sol. Like Anthropic’s Claude Cowork, ChatGPT Work is an “AI agent” that can run software applications and websites, including spreadsheets, online calendars, and email services, on behalf of users.
In standard benchmarks, OpenAI’s new flagship roughly matches the performance of Anthropic’s flagship, the Fable 5, according to Vals AI, a company that tracks the performance of the latest artificial intelligence technologies.
“It’s state-of-the-art in real-world situations involving financial and legal tasks,” said Rayan Krishnan, CEO of Vals AI.
The release of the more powerful model follows efforts by the US government to restrict both OpenAI and Anthropic’s new artificial intelligence models due to cybersecurity concerns. While President Trump has long taken a lax approach to regulating the AI industry, last month he signed an executive order asking tech companies to voluntarily provide government oversight of new AI models before they go public.
Last week, OpenAI said it would only share its new technology with a small group of companies approved by the Trump administration. However, after lengthy discussions, the administration allowed both OpenAI and Anthropic to make their new technologies public.
Although the administration initially treated the technologies as threats to national security, many independent security experts said the dangers were overstated.
OpenAI takes a more open approach to cybersecurity than Anthropic, which has strictly limited the use of Fable to prevent hackers from using its technology. The new OpenAI model includes fewer guardrails so that individuals and businesses can use the technology to defend themselves.
In April, Anthropic released a new artificial intelligence system called Claude Mythos, but said it could not share the technology with the public. Some executives and government officials feared the technology would usher in a new era of cyber threats.
Last month it released Fable 5, a more limited version of Mythos. Days later, the Trump administration ordered Anthropic to suspend access to all foreign nationals, citing national security concerns. To comply with the order, Anthropic completely disconnected the technology.
The company has added guardrails to Fable to block responses related to cybersecurity, biology, and other vulnerable areas. When Fable blocks responses, it sends them to a less powerful technology, Claude Opus 4.8.
Technologies like Fable and Sol are particularly good at identifying vulnerabilities in the software that underpins the Internet. This means that malicious hackers can use this technology to attack computer networks. But it also means businesses and governments can use it to defend against these attacks.
Fable’s guardrails can prevent hackers from using the technology for offensive purposes. But those same barriers can also prevent businesses and governments from using the system to defend themselves.
Like Anthropic’s Fable, OpenAI’s GPT-5.6 Sol is expensive compared to less powerful models. Fable is roughly twice as expensive as Anthropic’s next-highest-performing model, and OpenAI’s model is in some cases more expensive than Fable, Mr. Krishnan said.
On Thursday, Meta also released new AI technology. It wasn’t as powerful as the GPT-5.6 Sol or Fable, and it’s also significantly cheaper.
(The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to AI systems. Both companies have denied the claims.)