GPT-5.6 Series Preview: Meet Sol, Terra, Luna—OpenAI’s Next-Gen AI Models

As businesses and researchers demand AI tools that deliver unmatched power without compromising on cost or safety, OpenAI is stepping forward with a groundbreaking new lineup. Today, the company announced a limited preview of its GPT-5.6 Series—three distinct models designed to cater to every use case, from high-stakes scientific research to fast-paced, cost-effective workflows.

Introducing the GPT-5.6 Series: Three Models for Every Need

The GPT-5.6 Series marks a shift in OpenAI’s naming strategy: the numerical identifier denotes the model generation, while Sol, Terra, and Luna represent stable, independently evolving capability tiers. This structure lets users easily balance intelligence, speed, and cost based on their specific requirements:

1. Sol: The flagship model, boasting OpenAI’s most advanced capabilities to date.
2. Terra: A balanced option that matches the performance of GPT-5.5 at half the cost.
3. Luna: A fast, economical model that delivers powerful AI at the lowest price point in the series.

GPT-5.6 Sol: The Flagship Powerhouse

Sol stands as the pinnacle of the GPT-5.6 Series, packed with features that push the boundaries of what AI can accomplish:

1. Enhanced Reasoning: Introduces max reasoning intensity for deep, complex problem-solving, plus an ultra mode that leverages sub-agents to accelerate multi-step tasks.
2. Industry-Leading Coding: Achieved a new benchmark on Terminal-Bench 2.1, excelling at command-line workflows that require planning, iteration, and tool coordination.
3. Advanced Biological Capabilities: Outperformed GPT-5.5 on GeneBench v1 using fewer tokens, making it ideal for long-cycle genomics and quantitative biology analysis.
4. Unmatched Cybersecurity: OpenAI’s most secure model yet, advancing the performance-efficiency frontier for long-cycle security tasks like vulnerability research and exploit development. It matches Mythos Preview on ExploitBench using roughly one-third the output tokens, and across the GPT-5.6 Series, cybersecurity capabilities scale significantly with reasoning intensity on the UC Berkeley-developed ExploitGym benchmark.

Layered Safety: Balancing Power with Responsible Innovation

OpenAI’s core strategy for the GPT-5.6 Series is to advance AI capabilities while mitigating risks, ensuring legitimate defensive work (like code reviews and vulnerability research) remains accessible while restricting malicious use.

Key safety measures include:

1. Threshold Compliance: GPT-5.6 Sol does not cross the "critical" cybersecurity threshold, as tests confirmed it cannot autonomously generate full attack-chain exploits. A phased release approach addresses remaining uncertainties.
2. Layered Protection Stack: A multi-tiered system covering model-level training safeguards, real-time output checks (with classifier evaluations and secondary model reviews for high-risk cases), account-level audits to distinguish between malicious behavior and legitimate dual-use work, differentiated access, ongoing monitoring, and continuous testing.
3. Rigorous Red Team Testing: Over 700,000 A100-equivalent GPU hours were invested in automated red team testing focused on universal jailbreak attacks, complemented by third-party human expert testing. A rapid response process is in place to reproduce, evaluate, and fix newly discovered jailbreak vulnerabilities.

Release Rhythm: From Limited Preview to Wide Availability

The GPT-5.6 Series is currently in limited preview, accessible only to trusted partners. OpenAI has shared participant information with the U.S. government as part of collaborative efforts to develop cybersecurity executive order frameworks and repeatable release processes.

Looking ahead:

1. Universal availability for the entire GPT-5.6 Series is planned for the coming weeks.
2. Access will soon expand to ChatGPT, Codex, and API users.
3. In July, GPT-5.6 Sol will launch on Cerebras, offering speeds up to 750 tokens per second, initially limited to select clients.

Pricing & Cost-Saving Features

Pricing for the GPT-5.6 Series is structured per 1 million tokens, with clear differentiation across models:

1. Sol: $5 for input tokens, $30 for output tokens.
2. Terra: $2.50 for input tokens, $15 for output tokens.
3. Luna: $1 for input tokens, $6 for output tokens.

To help users reduce costs, OpenAI has introduced a new prompt caching feature. It supports explicit cache breakpoints with a minimum 30-minute lifecycle: cache writes are billed at 1.25x the uncached input rate, while cache reads receive a 90% discount.

With the GPT-5.6 Series, OpenAI is not just launching new models—it’s redefining how AI balances power, accessibility, and safety. By offering tailored options for every use case, implementing robust protective measures, and planning a transparent rollout, the company is paving the way for responsible AI innovation that benefits businesses, researchers, and society at large.