OpenAI unveils GPT-5.6 and ChatGPT Work to boost workplace productivity

OpenAI has officially released GPT-5.6 after staggering the model's rollout at the request of the U.S. government. The launch includes three variants: SOL, the company's most capable model; Luna, which is optimized for speed; and Terra, which offers a balance between performance and efficiency for everyday tasks.

The flagship SOL model also introduces an Ultra mode, allowing the system to devote more compute to complex tasks and delegate work across multiple sub-models. Speaking to CNBC, OpenAI CEO Sam Altman said the SOL variant is 54% more token-efficient on agentic coding tasks.

Alongside the new models, OpenAI has also introduced GPT Work, an AI-powered workspace built on GPT-5.6. The tool can gather context across connected applications and files to generate documents, spreadsheets, presentations, and other work-related content.

GPT Work will also be able to operate across the web, smartphones, and computers. The feature is rolling out first to the ChatGPT apps on macOS and Windows for users across all subscription tiers, with web support set to follow.

More from Tech

GPT-5.6 has been designed to improve end-to-end knowledge work by transforming information from documents and connected workplace tools into polished, shareable outputs. The model can understand and process context from platforms such as Slack, Notion, Microsoft 365, and Google Drive, enabling it to generate documents, reports, presentations, and other professional content with minimal user input.

According to OpenAI, GPT-5.6 also delivers improvements in web browsing, tool use, and computer-assisted workflows. The three model variants cater to different needs, with Sol targeting advanced professional tasks, Terra offering a balance of performance and affordability, and Luna focusing on speed and cost efficiency.

OpenAI GPT-5.6: New features, safety upgrades, availability and pricing

OpenAI has introduced GPT-5.6 with a range of new capabilities, stronger safety protections and a revised pricing structure. The model family is available in three variants—Sol, Terra, and Luna—each designed for different performance and cost requirements.

New features

GPT-5.6 is available across ChatGPT, ChatGPT Work, Codex, and the OpenAI API.

The flagship Sol model is designed for the most demanding reasoning and coding tasks, while Terra offers a lower-cost balance between performance and efficiency. Luna is the fastest and most affordable variant for lightweight workloads.

Developers can access all three models through the OpenAI API. The update introduces Programmatic Tool Calling, allowing GPT-5.6 to write and execute programs in memory while coordinating multiple tools. OpenAI has also launched Multi-agent (beta), enabling GPT-5.6 to run several AI sub-agents simultaneously and combine their outputs into a single response.

GPT-5.6 also brings improved prompt caching with explicit cache breakpoints and a minimum cache lifetime of 30 minutes, helping developers reduce latency and API costs.

Safety improvements

OpenAI says GPT-5.6 features its most advanced safety system to date, with multiple layers of protection including model-level safeguards, real-time monitoring, account-level enforcement and a new reasoning monitor that evaluates whether a request could cause harm before allowing advanced capabilities.

While GPT-5.6 is more capable than previous models in cybersecurity and biology, OpenAI says it does not cross the company's "Critical" risk threshold in either domain. The company added that the model is better suited for helping security researchers identify and fix software vulnerabilities than for carrying out autonomous cyberattacks.

Compared with previous models, OpenAI says GPT-5.6 Sol blocks roughly 10 times more potentially harmful cyber-related activity. Before launch, the company conducted its largest-ever safety evaluation, including external red teaming and approximately 700,000 A100e GPU hours of automated security testing.

Availability