This episode covers OpenAI's release of ChatGPT Work, a new agentic harness that extends Codex-like functionality to general knowledge work, alongside the official launch of the GPT-5.6 model family. The discussion highlights a major shift in AI competition from pure model performance to cost efficiency and harness capabilities, with Meta's Muse Spark 1.1 and Grok 4.5 also entering the frontier conversation.
Summarized by Podsumo
OpenAI launched ChatGPT Work, an agentic harness that brings Codex-style automation to knowledge work, with connectors for Notion, Google Drive, and Microsoft 365, and the ability to run tasks even when your laptop is closed.
The GPT-5.6 model family (Sol, Terra, Luna) was officially released, with benchmarks emphasizing cost efficiency—Sol matches or beats competitors at a fraction of the price, being 40% cheaper than Opus 4.8 and a third the cost of Fable 5.
Meta's Muse Spark 1.1 emerged as a surprise frontier contender, offering competitive performance at one-tenth the cost of Fable and GPT-5.5, with incredibly fast inference speeds.
The episode highlights a major shift in AI competition: labs are now competing on cost and harness capabilities, not just raw model performance, with Grok 4.5 and Muse Spark 1.1 re-entering the conversation.
OpenAI's ChatGPT Work extends Codex-style agentic automation to knowledge work, allowing users to define goals and let the system complete multi-step tasks across apps like Notion, Google Drive, and Microsoft 365.
"GPT-5.6 Sol is the first model I've trusted to run whole loops of knowledge work, not just help with individual tasks. It has shifted my job from doing the work to tending the system that does it."
"Fable has way more big model smell. But that means it's a skill in itself to get value out of it, and 99% of people are still not there yet. GPT-5.6 is almost as powerful, but it's easy to use, fast, and relatively cheap."
"The model is so cheap I almost don't believe it. In practice, we see it's one-tenth the cost of both Fable and GPT-5.5."