OpenAI has previewed GPT-5.6, its next model family, starting with Sol, its flagship model. The release is not broadly available yet, but it gives a signal that ChatGPT is heading towards deeper reasoning, more agentic work, stronger coding capabilities, and more careful safeguards around high-risk use cases.

Here are the key updates:

• A new model family: GPT-5.6 introduces three tiers: Sol for the strongest intelligence, Terra for everyday work, and Luna for fast, lower-cost use. This gives users and developers clearer choices between depth, speed, and cost.
• Deeper reasoning for complex work: GPT-5.6 introduces a new “max” reasoning effort, giving Sol more time to work through difficult tasks. OpenAI is also adding an “ultra” mode that uses subagents to handle more complex, multi-step work.
• Stronger coding and technical workflows: GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1, a benchmark focused on command-line tasks that require planning, iteration, and tool coordination.
• More powerful cyber capabilities: Sol is also OpenAI’s most capable model yet for cybersecurity. Because these capabilities can be dual-use, OpenAI is pairing the release with a layered safeguard system, including model-level refusals, real-time checks, account-level monitoring, and phased access.

At ​the request of the U.S. government​, GPT-5.6 is starting with trusted partner access through the API and Codex, with broader availability expected later.

That echoes what is now happening with Anthropic’s Fable 5 and Mythos 5, where after an earlier suspension tied to national-security concerns, the U.S. government has allowed Anthropic to restore Mythos 5 access for more than ​100 trusted U.S. organizations​.

For leaders, the important shift is that model capability, product availability, and workplace adoption are no longer moving on the same timeline. The most advanced models may exist, be tested, and shape the market before most users can access them.

Read more AI news after this: