OpenAI Has Released Its Latest Series of AI Models, CEO Sam Altman Says That They Increase Efficiency By More Than 50%

OpenAI CEO Sam Altman said the company’s latest flagship artificial intelligence model delivers a 54% improvement in token efficiency for agentic coding tasks, a gain that could significantly reduce costs for businesses deploying AI-powered software development tools.

The announcement comes as OpenAI broadens access to its new GPT-5.6 family of models following a limited rollout that was initially restricted while U.S. officials reviewed the technology.

The flagship model, known as Sol, is designed for advanced reasoning and coding, while two additional variants, Terra and Luna, target enterprise workloads and high-volume, lower-cost applications.

Altman told CNBC that the improvement centers on “agentic coding,” an increasingly important category of AI in which models can independently code software with minimal human supervision. Rather than acting as a simple autocomplete tool, these AI agents are designed to carry out complex programming tasks across multiple files and workflows.

The 54% gain in token efficiency means the model requires substantially fewer tokens, the units AI systems use to process and generate text, to complete coding assignments. Lower token usage translates into reduced computing costs, faster execution and improved scalability for enterprise customers running large numbers of AI coding tasks.

“Every enterprise now is thinking about spend and the value they’re getting in exchange for AI, and this is what we really want to do,” Altman told CNBC.

OpenAI’s latest release arrives amid intensifying competition from rivals including Anthropic, Google and Elon Musk’s AI company SpaceXAI, which this week introduced its own new model focused on coding and agentic tasks. Companies are racing not only to build smarter AI systems but also to lower operating costs enough to encourage widespread enterprise adoption.

The rollout of GPT-5.6 was also notable because it underwent an unusually close government review before becoming broadly available. According to Altman, the Trump administration requested a staggered launch while federal agencies evaluated the model’s capabilities and safety claims. The review involved officials including Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent and U.S. National Cyber Director Sean Cairncross.

Altman characterized the process as collaborative, saying OpenAI worked with federal reviewers to address questions before receiving approval for a wider public release.
“If you want broad access, which we do, and you have powerful models, you really want to be able to be confident in your safety claims, because otherwise the world is going to get uncomfortable very fast,” Altman told CNBC.

During the review period, GPT-5.6 was available only to about 20 government-vetted partners, allowing regulators to examine its capabilities in areas including coding, cybersecurity and biology before a broader rollout.