Introduction
The GPT‑5.6 series marks a significant breakthrough in the field of artificial intelligence. Composed of the GPT‑5.6 Sol, Terra, and Luna models, this new generation addresses the diverse needs of users while integrating reinforced security features. With increased capacity and competitive pricing, these models promise to improve processes in key areas such as programming, biology, and cybersecurity.
The GPT‑5.6 Models: Sol, Terra and Luna
The new GPT‑5.6 series consists of three main models:
- GPT‑5.6 Sol: The flagship model, focused on advanced performance and maximum capabilities. It integrates an ultra mode to accelerate complex work and deepen reasoning.
- GPT‑5.6 Terra: A balanced model for everyday tasks, offering performance similar to GPT‑5.5 while being offered at half the cost.
- GPT‑5.6 Luna: Designed to provide solid capabilities at some of the lowest prices on the market.
According to OpenAI, the launch of GPT‑5.6 Sol comes with the most robust security framework ever deployed. Developers have improved protections to minimize risks associated with sensitive cyber requests, while strengthening the system against potential attacks through intensive testing and rigorous evaluation by third-party experts.
[IMAGE:index:https://openai.com/images/gpt5.6-launch:Performance-Highlights]
Exceptional Benchmark Performance
Improvements in Programming
With the Terminal‑Bench 2.1 benchmark, the GPT‑5.6 Sol model sets a new standard for command-line workflows, including planning, iteration, and coordination with tools. Thanks to its ultra mode and multi-agent capabilities, it delivers significantly better results than its predecessors.
Progress in Biology
On GeneBench v1, a benchmark dedicated to genomics and complex biological analyses, GPT‑5.6 Sol surpasses GPT‑5.5 by achieving better results while using fewer tokens. This makes it a powerful tool for quantitative biology analysis.
Advances in Cybersecurity
The model also stands out in the crucial field of cybersecurity. On ExploitBench², it excels in vulnerability detection and achieves performance comparable to the Mythos preview version, while using a third of the tokens required. A study on ExploitGym, conducted by UC Berkeley in partnership with OpenAI, also highlights the significant progress made by all GPT‑5.6 models in long-term security tasks.
Good to know
GPT‑5.6 Sol introduces a maximum reasoning mode, allowing for deeper analysis and accelerating complex tasks.
A Multi-Layered Security Architecture
Reinforced Safeguards
The GPT‑5.6 models benefit from multi-layered protections to address challenges posed by cybersecurity attacks. Among the measures implemented:
- Model-integrated filters: These algorithms prevent prohibited cybernetic assistance and detect bypass attempts.
- Real-time classifiers: These tools interrupt content generation when a potential violation is detected.
- Account-level review: In-depth analysis of conversations allows identification of persistent malicious activities.
These mechanisms ensure that the advanced capabilities of the models remain available for legitimate uses while making malicious use more difficult and detectable.
Caution
Certain protections could introduce delays or block legitimate requests, particularly in domains where defensive and offensive uses overlap.
Intensive Testing and Rapid Redevelopment
To ensure the effectiveness of protections against evolving adversarial tactics, OpenAI dedicated over 700,000 A100 GPU hours to automated red-teaming. Third-party experts also participated in robustness testing, simulating jailbreak campaigns specially designed to bypass protections.
Through this process, system weaknesses are detected more quickly and addressed effectively, ensuring better adaptation to real-world scenarios and enhanced security levels.
Availability and Pricing
Flexible Pricing by Model
The GPT‑5.6 models are offered with simplified pricing per million tokens:
| Model | Input Price (USD) | Output Price (USD) |
|---|---|---|
| Sol | $5 | $30 |
| Terra | $2.50 | $15 |
| Luna | $1 | $6 |
These prices also include new features such as a more predictable caching system, with a minimum lifetime of 30 minutes for each cache.
Initial Access and Progressive Expansion
The GPT‑5.6 models will initially be available via the API and Codex for a limited group of trusted partners. Access will expand soon to ChatGPT users and other enterprises.
[IMAGE:index:https://openai.com/images/pricing-gpt5.6:Pricing-Table]
Tip
For users needing reduced processing time, OpenAI offers a processing priority option in the API and a fast mode in Codex.
Conclusion
With the GPT‑5.6 series, OpenAI reaches a new horizon in artificial intelligence, combining optimal performance and advanced security. These advances are promising for enterprises, developers, and researchers. The preview phase will allow for refinement of the experience before broader deployment.
For those seeking to optimize their operations by leveraging a powerful AI model, GPT‑5.6 could be the ideal solution for meeting the complex requirements of modern technology.
