GPT-5.6 Sol Launches After Commerce Clearance

OpenAI has released its new flagship model, GPT-5.6 Sol, to the public. This follows a two-week preview where access was limited to about 20 trusted partners by the U.S. Department of Commerce. The launch also introduces two smaller models named Terra and Luna. It is the first time the company has used names instead of numbers for its tiers.

The new naming strategy sets three distinct capability levels that update on their own schedules. Sol acts as the top-tier model, while Terra serves as an everyday option matching GPT-5.5 performance at half the price. Luna stands as the most affordable entry point. Pricing for Sol sits between 5 dollars and 30 dollars per million tokens for input and output respectively.

This places it above Chinese low-cost challengers like DeepSeek but below premium U.S. rivals such as Anthropic's Claude Fable 5, which charges 10 to 50 dollars. Sol introduces two new controls: a max reasoning knob that lets the model think longer and an ultra mode designed for subagent delegation. In benchmarks, Sol in ultra mode scored 91.9 percent on Terminal-Bench 2.1, beating Anthropic's Claude Mythos 5 at 88.0 percent.

On ExploitBench, a test of finding software vulnerabilities, it matched the restricted Mythos Preview while using roughly one-third fewer tokens. Despite these gains, Sol remains below the "Cyber Critical" line in OpenAI's own risk framework. Early reactions from developers have been mixed but generally positive for computer use tasks. Theo, a prominent developer and CEO of T3 Chat, called Sol world leading for interacting with computers.

However, researcher Daichi Konno observed that Anthropic still leads in writing tasks. The launch coincides with a shift at Anthropic, where its Fable 5 model was removed from standard subscriptions and moved to usage credits only after its weekly allowance expired on July 7. This rollout highlights how crowded the frontier has become, with OpenAI, Anthropic, and SpaceX now holding valuations exceeding all U.S. tech exits since 2000 combined.

Meta is entering this coding market with Muse Spark 1.1 to chase the leaders, while Google faces pressure as its Gemini 3 remains the oldest flagship model still active since November 2025. Leaked roadmaps suggest Sol's dominance may be short-lived, as GPT-5.6 appears to be the final 5.x iteration before a larger GPT-6 base arrives within a month. Watch whether Sol maintains its edge as these next-generation competitors hit the market in the coming weeks.