OpenAI’s Strongest GPT-5.6 Models Arrive Behind A Locked Door

OpenAI previewed its three-model GPT-5.6 family Friday, pairing sharp gains in coding, biology and cybersecurity with access limited for now to a small circle of partners.

Key Points:

GPT-5.6 Family Targets Coding And Cyber Work

The company unveiled Sol as its strongest model yet, Terra as a balanced everyday tier priced at roughly half its predecessor, and Luna as the fastest option at $1 per million input tokens. A fresh naming scheme now uses the number to mark a model's generation, while the names stand for durable capability tiers. Each tier can advance on its own schedule, giving developers a clearer choice across speed, depth and price.

GPT-5.6 also debuts two ways to push Sol harder on long, multi-step jobs. A max setting buys the model more thinking time, while an ultra mode splits complex work across subagents that run in parallel before merging their results.

OpenAI also plans to run Sol on Cerebras chips at up to 750 tokens per second in July.

Sol set a new high mark on Terminal-Bench 2.1, a test of command-line work that demands planning and tool coordination. In genomics analysis, it edged past GPT-5.5 while spending fewer tokens.

On one exploit benchmark, Sol matched Anthropic's Mythos Preview using roughly a third of the output tokens.

Also Read: Chainlink's Wallet Record Turns LINK's $9 Rebound Into The Main Test

ChatGPT Risk Tier Reshapes AI Access

OpenAI classified all three models as high risk for both cyber and biological misuse, a first for its smaller, cheaper tiers. None of them, the firm said, reached the critical level that would trigger its tightest restrictions, and Sol proved better at finding and fixing flaws than running full attacks.

That rating helps explain why access stays narrow for now. Roughly 20 organizations cleared by the government can run the models during the preview, with a wider rollout to ChatGPT, Codex and the API promised within weeks. Security researcher Alex Stamos warned that walling off frontier tools could hand rivals abroad an edge in the AI race.

The launch caps a year of tightening cyber controls at the company, from the first safeguards on GPT-5.2 last December to stricter classifiers on GPT-5.5. For GPT-5.6, OpenAI says it spent more than 700,000 A100-equivalent GPU hours on automated red-teaming alone.

Read Next: XRP Falls Near $1 While ETF Buyers Test A Weak Spot Market