OpenAI releases GPT-5.6 Sol with 54% token efficiency gain on agentic coding tasks

OpenAI released its GPT-5.6 series models—Sol, Terra, and Luna—for broad availability on Thursday, following an initial limited launch requested by the U.S. government. CEO Sam Altman reported that GPT-5.6 Sol achieves 54% greater token efficiency on agentic coding tasks compared to previous versions.

"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. He claims the models perform "as good or better" than competing models currently available.

Government safety review process

The company worked with Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and U.S. National Cyber Director Sean Cairncross on the approval process before broad release. OpenAI initially limited deployment to "a small group of trusted partners" at government request.

Altman described the process as a "collaborative back and forth," where government officials conducted tests and raised safety concerns for OpenAI to address. "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," he said.

Technical specifications not disclosed

OpenAI has not yet released detailed technical specifications for the GPT-5.6 series, including context window size, parameter count, benchmark scores, or pricing structure. The company announced the models last month but provided limited technical information at that time.

The 54% token efficiency improvement specifically applies to agentic coding tasks—autonomous programming workflows where AI agents write, test, and debug code with minimal human intervention. Token efficiency directly impacts operational costs for enterprise deployments, as most AI providers charge based on token usage.

What this means

The token efficiency gain addresses a critical enterprise concern: AI deployment costs. A 54% reduction in tokens required for agentic coding tasks translates directly to cost savings for companies running autonomous coding workflows at scale. However, without published benchmarks, pricing, or independent verification, the practical impact remains unclear. The government safety review process signals increased regulatory scrutiny for powerful AI models, potentially establishing a precedent for future releases from other labs.