This video covers the latest improvements in the AI model version 2.7, comparing it directly to its predecessor, 2.5. If you have been following the channel for a while, you might remember the early days of M2.1, when the AI agent — nicknamed Jeff — would pop up with a casual greeting and little else. Now, the conversation quality has evolved significantly, offering more personality and meaningful engagement, which matters if you care about the experience of working with your AI agents day to day. On the practical side, the team's early testing of 2.7 reveals several notable upgrades. First, the model is less prone to overthinking. Where 2.5 would sometimes spiral when handling complex tasks, 2.7 is more decisive and cuts through ambiguity faster. Second, tool calling has improved. Version 2.5 already ranked among the best with a 76.8% score on the BFCL benchmark, and 2.7 looks set to push that number even higher. Third, instruction following is tighter — if you give the model a structured plan or step-by-step instructions, it sticks to them reliably, reducing the need for repeated corrections. Speed is another area where 2.7 holds up well. The video contrasts it with other models that can take 10 to 15 minutes to complete tasks like building a presentation. With 2.7, the same kind of output takes roughly two to three minutes, making it a noticeably faster option for real workflows. Overall, if you are using AI agents for productivity tasks, 2.7 represents a meaningful step forward — not just in raw capability, but in how reliably and efficiently it executes your instructions.
No transcript available
No transcript available