OpenAI has released its new flagship AI model, GPT-5.6 Sol, to the public. The launch ends a two-week restricted preview held by the U.S. Department of Commerce.
Sol is accompanied by two smaller models, Terra and Luna. This marks a shift in OpenAI's naming strategy, moving from numbers to names to denote capability tiers.
On performance benchmarks, Sol in its ultra configuration scored 91.9% on Terminal-Bench 2.1, outperforming models from Anthropic and Google. It also matched Anthropic's restricted Mythos Preview on the ExploitBench cybersecurity test while using roughly a third of the computational tokens.
Early access users have praised Sol. Developer Theo called it "world leading in computer use," while Dan Shipper of Every described it as a "Porsche" for daily work compared to Anthropic's "warp drive" Fable model.
The launch intensifies competition. It arrives the same week Anthropic's Fable 5 moved to usage-only credits, following recent releases from xAI's Grok 4.5 and Meta's Muse Spark 1.1. Google's Gemini 3 remains the oldest flagship in the current cycle.