OpenAI began a limited preview of its GPT-5.6 model family on June 26, 2026, introducing three tiers: Sol, the flagship model targeting complex coding, biology, and cybersecurity tasks; Terra, positioned as a cost-efficient middle tier; and Luna, designed for high-volume, latency-sensitive workloads. The company said broader availability through ChatGPT and its API is expected in the coming weeks.
Sol scored 88.8% on Terminal-Bench 2.1, narrowly edging Claude Mythos 5 at 88.0%, while the new "ultra" mode—which deploys subagents to divide complex tasks—pushed that score to 91.9%. On GeneBench v1, a biology workflow benchmark, Sol outperformed GPT-5.5 by approximately nine percentage points. Pricing was set at $5/$30 per million input/output tokens for Sol and $1/$6 for Luna, with Terra priced at roughly half of GPT-5.5.
The release comes as frontier AI labs race to extend model capabilities into autonomous, multi-step workflows. OpenAI said it logged more than 700,000 A100-equivalent GPU hours on automated red-teaming during development. On cybersecurity benchmarks, Sol demonstrated the ability to identify vulnerabilities and exploit primitives but did not independently produce full end-to-end exploits in controlled testing on Chromium and Firefox environments.
Initial access is limited to approximately 20 companies whose participation has been shared with the U.S. government—a model OpenAI described as a temporary measure while broader AI policy frameworks are finalized. The company plans to bring Sol to Cerebras in July, targeting throughput of up to 750 tokens per second, with capacity expected to ramp gradually.