OpenAI Astra’s reasoning capabilities raise questions about chain-of-thought fidelity

OpenAI's latest model, Astra, demonstrates a significant increase in chain-of-thought controllability, but researchers note that the model can generate irrelevant reasoning while still providing correct answers. This suggests that reasoning fidelity depends on necessity rather than the model's honesty. Reports indicate that Astra's ability to solve tasks without visible reasoning has improved significantly compared to its predecessors.