OpenAI releases GPT-5.6 Sol, Terra, Luna after government safety review; Sol hits 91.9% Terminal-Bench 2.1

<cite index="13-2">OpenAI's advanced GPT-5.6 series artificial intelligence models became publicly available Thursday (July 9, 2026) following a limited preview period in late June.</cite> <cite index="14-2">Sol Ultra leads Terminal-Bench 2.1 at 91.9% on command-line workflows, exceeding Anthropic's Claude Mythos 5 (88.0%), while Terra matches Fable 5 on complex reasoning at half the cost.</cite> The three-tier family comes in Sol (frontier), Terra (cost-optimized everyday), and Luna (fastest/cheapest). <cite index="13-2">GPT-5.6 Sol is the strongest model of the series and is tuned for work in biology, chemistry and cybersecurity.</cite>

<cite index="11-5">GPT-5.6 Sol is OpenAI's most capable model yet for cybersecurity, shifting the performance-efficiency frontier for long-horizon security tasks including vulnerability research and exploitation. On ExploitBench, GPT-5.6 Sol is competitive with Mythos Preview using only ~1/3 of the output tokens.</cite> <cite index="15-3">On SecureBio evaluations, GPT-5.6 reached top reported scores including 53.5% on the Virology Capabilities Test, 60.0% on Molecular Biology, and 68.4% on Human Pathogen Capabilities—about 9 percentage points above GPT-5.5.</cite>

<cite index="12-3">The government-gated preview kept Sol, Terra, and Luna available only through the OpenAI API and Codex to a select group of trusted partners and organizations.</cite> <cite index="12-1">Separately, OpenAI is launching GPT-5.6 Sol on Cerebras hardware at up to 750 tokens per second in July 2026, initially limited to select customers as capacity expands.</cite> Pricing: Sol $5/$30 in/out, Terra $2.50/$15, Luna $1/$6 per 1M tokens. Architects: the government safety review precedent is significant—future frontier model launches may face similar gating. The shift to multi-tier pricing (Sol–Luna cost ratio: 5x) pushes teams toward model-routing architectures rather than default-to-best-model patterns.