GPT 5.6 is a highly advanced and efficient AI model capable of complex tasks such as building an Excel clone and a detailed Minecraft-like 3D world, with strong performance in enterprise applications and cost-effective pricing. Available in three sizes with flexible deployment options, it excels in automation and workflow optimization, though newer models like Fable hint at even greater future AI capabilities.
GPT 5.6 represents a significant advancement over its predecessor, GPT 5.5, delivering one of the most capable and efficient AI models to date. The creator demonstrated its power by setting a simple goal: to build an Excel clone with feature parity. Running continuously for over five days, GPT 5.6, combined with Codex, produced a highly functional Excel-like application featuring sorting, formulas, data validation, conditional formatting, tables, and pivot tables. This impressive feat highlights the model’s ability to handle complex tasks and replicate sophisticated software functionalities within a condensed, user-friendly HTML app.
One of the standout capabilities of GPT 5.6 is its exceptional proficiency in browser and computer use. The model can interact seamlessly with desktop applications, such as opening and manipulating Excel, and perform complex web-based tasks like managing Gmail or DNS records with simple prompts. This enhanced tool use makes GPT 5.6 a powerful assistant for automating and streamlining workflows. Additionally, the model was tasked with creating a Minecraft clone, which it developed over seven days, achieving a highly detailed and interactive 3D world with realistic features, mobs, and biomes, showcasing its ability to generate immersive virtual environments.
Benchmark tests conducted by Box AI reveal that GPT 5.6 performs competitively on enterprise-grade tasks involving document reading, number reconciliation, due diligence, and error review. While GPT 5.5 and other models like Terra and Luna showed varying accuracy levels, GPT 5.6 Luna matched Terra’s performance but with faster processing and lower costs. The pricing for GPT 5.6 is notably more economical, charging less per million input and output tokens compared to alternatives like Fable, making it a cost-effective choice for users seeking high-quality AI performance.
Despite GPT 5.6’s strengths, the newer Fable model offers a glimpse into the future of AI capabilities. Fable feels like a fresh, unoptimized model with greater potential, akin to a high-performance Ferrari straight off the production line, whereas GPT 5.6 is likened to a finely tuned Honda Civic that has been pushed to its limits. This distinction underscores that while GPT 5.6 is the pinnacle of its generation, Fable represents the beginning of a new era with even more advanced reasoning and problem-solving abilities yet to be fully realized.
GPT 5.6 is available in three model sizes—Luna (smallest), Terra (medium), and Sol (largest)—each offering different levels of reasoning and computational power. The Sol model can operate at an Ultra level for intensive tasks, allowing users to optimize performance and cost by delegating tasks across models based on complexity. This flexibility supports sophisticated workflows where planning, implementation, and deployment can be distributed efficiently. The creator also developed a skill for Codex to automate this delegation, enabling users to maximize quality and minimize resource consumption, further enhancing GPT 5.6’s practical utility in real-world applications.