OpenAI previews GPT-5.6 to select partners with three variants priced from $1 to $30 per million tokens

OpenAI has begun previewing its GPT-5.6 series to a limited group of trusted partners after providing early access to the US government. The company plans a broader public release in the coming weeks.

Three variants with distinct pricing

The GPT-5.6 series includes three models:

• Sol: $5 per million input tokens, $30 per million output tokens. OpenAI claims this is its strongest model yet with enhanced cybersecurity capabilities.
• Terra: $2.50 input/$15 output per million tokens. According to OpenAI, Terra matches GPT-5.5 performance at half the cost for everyday use.
• Luna: $1 input/$6 output per million tokens. The lowest-cost option in the series.

For comparison, Anthropic's suspended Fable 5 model cost $10 input/$50 output per million tokens before access was blocked.

Government review and cybersecurity focus

The limited preview follows President Trump's AI cybersecurity order from earlier this month, which requests voluntary government review of powerful AI models 30 days before public release. According to The New York Times, OpenAI, Anthropic, Google, xAI, and Microsoft had been providing early government access even before the executive order. Meta reportedly has not participated.

OpenAI stated it is taking this "short-term step" to ensure it can release the models publicly soon, but added: "We don't believe this kind of government access process should become the long-term default."

Technical capabilities and safeguards

GPT-5.6 Sol introduces a "max" reasoning effort mode, giving the model more time for deep reasoning tasks. OpenAI claims Sol is its most capable model for finding and fixing security vulnerabilities, with strengthened protections for high-risk activities.

The company reports spending 700,000 GPU hours identifying universal jailbreaks to develop countermeasures. All three variants include training to refuse "prohibited cyber assistance" and jailbreak attempts. OpenAI committed to implementing a "rapid-response process to reproduce, assess, prioritize, and remediate newly discovered jailbreaks."

Context: Anthropic's recent suspension

The cybersecurity emphasis follows Anthropic's suspension of its Mythos 5 and Fable 5 models two weeks ago after government directive. Amazon and other companies had reportedly notified authorities that those models could be jailbroken for malicious use. The US government has since given Anthropic permission to redeploy Mythos to select organizations.

What this means

The GPT-5.6 pricing positions OpenAI competitively below Anthropic's previous Fable 5 rates, while the three-tier structure allows developers to optimize for cost versus capability. The government review process and extensive jailbreak prevention measures signal that frontier AI labs are now operating under increased regulatory scrutiny, particularly for models with potential cybersecurity applications. Whether this voluntary review process becomes mandatory remains unclear, but the participation of major labs suggests industry acceptance of government oversight for the most capable models.