Open models hit coding parity, frontier reasoning leads

The "frontier" is no longer a fixed destination but a moving target that proprietary labs defend with massive compute and multi-step reasoning moats.

• –Open models like DeepSeek and Qwen have achieved functional parity in 70-80% of developer use cases, specifically coding and summarization.
• –The remaining gap persists in "deep multi-step reasoning" and broad factual accuracy under high ambiguity.
• –Economic pressure is mounting as high-tier local models now run competitively on consumer hardware, challenging the API-first business model.
• –Proprietary models are increasingly reserved for the absolute ceiling of complexity, while open weights become the production default.