GPT-5.6 Sol starts limited preview with cyber guardrails and $5 input pricing

The GPT-5.6 rollout puts deployment controls, security safeguards, and pricing on the same page as benchmark gains. OpenAI opened a limited preview on June 26, 2026 for three tiers: Sol as the flagship model, Terra as the balanced option, and Luna as the lower-cost model. Initial access runs through the API and Codex for selected trusted partners and organizations, with broader ChatGPT, Codex, and API availability planned later.

The capability claims are concrete enough to matter for developers. OpenAI says GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1, which tests command-line workflows that require planning, iteration, and tool coordination. The company also says Sol improves biology workflows on GeneBench v1 versus GPT-5.5 while using fewer tokens. In cybersecurity, Sol is described as competitive with Mythos Preview on ExploitBench while using about one-third of the output tokens.

The release is constrained because those same cyber gains increase dual-use risk. OpenAI says the models were previewed with the U.S. government before launch, and the first trusted-partner group has been shared with the government. The company says this should not become the long-term default, but frames it as a short-term path toward wider availability while a cyber Executive Order framework is being developed.

The safety stack is also part of the product. GPT-5.6 uses model-level refusals, real-time cyber and biology misuse classifiers, account-level review signals, differentiated access, monitoring, and continued red teaming. OpenAI says it spent more than 700,000 A100-equivalent GPU hours on automated red teaming for universal jailbreaks, then paired that work with third-party human expert testing.

Pricing gives teams a first budget line. Sol costs $5 per 1M input tokens and $30 per 1M output tokens; Terra is $2.50/$15; Luna is $1/$6. Cache writes are billed at 1.25x the uncached input rate, while cache reads keep the 90% cached-input discount. OpenAI also plans to launch GPT-5.6 Sol on Cerebras at up to 750 tokens per second in July. The primary source is OpenAI’s GPT-5.6 Sol preview.