Early testers say OpenAI's Astra 6.2 can stop partway through coding projects and produce written output cut so short that it is hard to follow. The complaints have drawn attention because they target the developer audience that AI companies have courted most closely.
OpenAI's case rests on cost. On one benchmark, the cost per task has fallen 35%, which matters to businesses running large volumes of work through a model. The episode asks whether the company is betting that cheaper, faster output for millions of office workers and routine busywork counts for more than winning over developers. Whether that bet pays off depends on whether the model is reliable enough for everyday tasks, and early testing has not settled that.
The same episode also covers Nvidia's board saying it will not reinstate Jensen Huang, and a wave of outages that has renewed debate over the rush to modernise ageing systems. For Australian businesses and professionals, the practical picture is mixed. Lower per-task cost could reduce spending on routine office work, but a single benchmark is not a guarantee, so teams should test the model on their own tasks before changing budgets. Developers should check that long coding outputs are complete before relying on them. If staff paste client or employee details into the tool, the Privacy Act 1988 applies to how that personal information is handled, so review data settings and vendor terms first. For most people this is a low-risk trial for drafting work with a human reviewing the result, not a reason to switch tools.
Key Points
Early testers report Astra 6.2 stops partway through coding projects and cuts off its writing.
OpenAI's cost per task is down 35% on one benchmark.
The release appears aimed at routine office work more than at developers.
The episode also flags Nvidia's board refusing to reinstate Jensen Huang and outages tied to aging systems.