OpenAI Launches GPT-5.6: A New Generation of AI Models

On June 27, OpenAI officially introduced GPT-5.6, advancing its large language model technology. This release consists of a full series with three distinct variants, designed to address diverse application needs and budget considerations.

Three Models, Three Strategic Positions

The GPT-5.6 series is built around three core versions, each optimized for specific use cases.

  • Sol (Flagship): Positioned as the top-tier model for agentic tasks, representing the peak performance of the series.
  • Terra (Balanced): Delivers performance close to GPT-5.5 but at approximately half the cost, striking an efficient balance between capability and expense.
  • Luna (Efficient): Focused on high throughput and low cost, suitable for large-scale processing with constrained budgets.

Transparent Pricing Structure

OpenAI has adopted a clear usage-based pricing model, with costs calculated per million input and output tokens. Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra at $2.5/$15; and Luna at $1/$6. This tiered approach allows developers to select the most economical option for their specific needs.

Benchmark Performance Highlights

In the Terminal-Bench 2.1 evaluation, the Sol variant achieved an overall score of 88.8%, with its Ultra version reaching 91.9%. Notably, on ExploitBench cybersecurity tasks, the new models demonstrated the ability to accomplish complex tasks using fewer tokens, indicating improved reasoning efficiency.

Enhanced Security Measures

OpenAI emphasized GPT-5.6's security features, describing them as the "strongest security stack to date." The security system underwent red team testing equivalent to 700,000 hours of A100 GPU computation, where researchers simulated various attack scenarios to identify and address potential vulnerabilities.

Limited Access and Rollout Timeline

Due to compliance with U.S. government export regulations, GPT-5.6 is currently available only to trusted partners via Codex and API channels. OpenAI stated it is actively working to meet regulatory requirements and plans to gradually expand access over the coming weeks, aiming for a full public release.