The video discusses the recent public release of OpenAI’s GPT 5.6 family of models, which includes three versions named Luna, Terra, and the flagship Gigabrain Sol. This release follows government approval after an executive order requiring AI labs to voluntarily submit their most powerful models for review before public deployment. GPT 5.6 Sol is highlighted for its impressive performance, particularly in agentic coding tasks, where it can spawn multiple sub-agents to work in parallel, significantly enhancing productivity and problem-solving capabilities.
OpenAI’s new approach with GPT 5.6 focuses less on making the base model smarter and more on leveraging a virtual workforce of AI sub-agents. These sub-agents can handle different programming tasks simultaneously, such as writing React components, managing databases, or designing user interfaces. This multi-agent orchestration effectively renders many existing startups in this space obsolete. The model features two key settings: Max Reasoning, which enhances deep thinking, and Ultra Mode, which activates the sub-agent army for complex tasks.
Benchmark tests reveal that GPT 5.6 Sol performs exceptionally well, particularly on Terminal Bench 2.1, a benchmark that evaluates real command-line workflows. Sol in Ultra Mode achieves a high score of 91.9%, outperforming competitors like Claude Mythos 5. However, it slightly underperforms Claude Mythos in cybersecurity benchmarks. Notably, OpenAI did not publish scores for the Swebbench Pro benchmark, where Anthropic’s Fable 5 currently leads, suggesting potential underperformance by GPT 5.6 in that area. Additionally, some concerns about GPT 5.6’s evaluation methods were raised due to detected shortcuts in testing.
The video compares GPT 5.6 Sol with Anthropic’s Claude Fable 5, noting that both are highly intelligent but serve different user needs. Sol is described as a cost-effective and fast “contractor” that completes tasks quickly with a team of sub-agents, while Fable is likened to a slower but meticulous contractor who charges more. The choice between the two depends on the specific requirements of the user, emphasizing that both models represent significant advancements in AI capabilities.
Finally, the video is sponsored by Blacksmith, a service that accelerates GitHub Actions by running them on high-performance bare metal gaming CPUs, reducing costs and improving observability in CI pipelines. This is particularly useful when working with AI agents generating large amounts of code autonomously. The sponsor offer includes 3,000 free minutes per month for new users, providing a practical tool for developers leveraging these advanced AI models. The video concludes by encouraging viewers to try Blacksmith and stay tuned for more updates on AI developments.
