OpenAI Unveils GPT-5.6: Luna, Terra, and Sol Revolutionize AI

⚡

Key Takeaways

1OpenAI launches GPT-5.6 with three models: Luna, Terra, and Sol, each with distinct capabilities.

2Pricing ranges from $1 to $30 per million tokens, depending on the size and power of the chosen model.

3GPT-5.6 outperforms Claude Fable 5 in 55 professional domains, setting a new performance standard.

💡Why it matters — This advancement strengthens OpenAI's position in the AI field, providing more efficient and affordable solutions for businesses.

OpenAI Introduces GPT-5.6: Luna, Terra, and Sol

OpenAI has recently launched its latest flagship model, GPT-5.6, which comes in three distinct versions: Luna, Terra, and Sol. These models are designed to meet various information processing needs, ranging from the simplest to the most complex tasks.

Pricing and Comparison with Competitors

The new GPT-5.6 models are offered at rates based on one million tokens of input and output. Luna, the most compact model, is available for $1 for input and $6 for output. Terra, the intermediate model, costs $2.50 for input and $15 for output. Finally, Sol, the most advanced model, is priced at $5 for input and $30 for output. To put these prices in perspective, the Claude Opus series is offered at $5 for input and $25 for output, while Claude Fable 5 is priced at $10 for input and $50 for output. However, these figures do not provide a complete picture, as the number of tokens required to accomplish a task can vary significantly between models.

Enhanced Capabilities and Performance

All models in the GPT-5.6 series have a knowledge cutoff date set for February 16, 2026. They offer an impressive context window of one million tokens and can generate up to 128,000 tokens of output. OpenAI highlights the exceptional performance of these models in long-term tasks. In a benchmark assessing professional workflows across 55 different domains, the GPT-5.6 Sol model achieved a record score of 53.6, surpassing Claude Fable 5 by 13.1 points. Even with average-level reasoning, GPT-5.6 Sol outperforms Claude Fable 5 by 11.4 points while being more cost-effective. Additionally, the smaller models, GPT-5.6 Terra and GPT-5.6 Luna, also outperform Fable 5 for about one-sixteenth of the cost.

Benchmarks and Challenges

Despite these successes, GPT-5.6 has not dominated all benchmarks. For instance, in the SWE-Bench Pro test, Claude Fable 5 scored 80%, surpassing the GPT-5.6 Sol model, which achieved 64.6%. OpenAI acknowledged this result and advised model developers to carefully review the SWE-Bench Pro results, estimating that 30% of the tasks in this benchmark could be flawed.

User Experience and New Features

Having had the opportunity to test GPT-5.6 Sol in advance, I found that the model is highly capable, although it does not yet surpass Claude Fable 5 for complex coding tasks. OpenAI has also introduced several interesting new API features, such as programmatic tool calling, which allows models to compose and execute JavaScript to orchestrate tool calls. This feature could bridge the gap between personal computing models and full terminal sessions.

API Innovations and Cost Optimization

Other innovations include multi-agent capability, which allows the model to launch sub-agents for parallel and targeted work, and prompt cache breakpoints, which offer increased control over cache breakpoints. These features are designed to optimize costs and improve the efficiency of the models.

Costs and Demonstrations

A dedicated page presents 18 different reasoning scenarios with the three models, illustrating their token costs. For example, the gpt-5.6-luna model at zero effort costs 0.71 cents, while the gpt-5.6-sol model at maximum reasoning level costs 48.55 cents. In a live demonstration, OpenAI showcased 3D pelicans using various means of transportation, illustrating the creative and technical capabilities of their technology.