OpenAI's lead under pressure as rivals start to close the gap (www.ft.com)

OpenAI’s dominance in generative AI is increasingly being challenged as competitors and the open-source community narrow the capability and deployment gap. Over the past year, multiple players — from well-funded rivals to collaborative OSS projects — have shipped models and products that match or approach OpenAI on many benchmark tasks, offer comparable multimodal capabilities, or undercut on price and customization for enterprises. The result is more choice for developers and businesses: multi-vendor strategies, self-hosting options, and specialty models tuned for specific domains are becoming practical alternatives to a single-cloud, API-first approach. Technically, the squeeze comes from a mix of algorithmic and systems advances: instruction tuning and reinforcement learning from human feedback (or “constitutional” approaches) have improved alignment; retrieval-augmented generation, adapters and parameter-efficient fine-tuning enable domain specialization without massive compute; and distillation, quantization and pruning cut inference cost and latency. Hardware and software co-design (new accelerators, memory-optimized runtimes) further lower deployment barriers. For the AI community this means faster iteration cycles, downward pressure on pricing, and a more diverse ecosystem — but also amplified challenges for safety, governance and standard-setting as more capable models proliferate outside a single steward.