Day 1 impressions of Fable 5.1:
• talks much more like a normal human now, less Claudish. • web search still feels worse than GPT-5.6 Sol. It often misses the key things to search for. • still much better than GPT models at frontend. • my daily driver is now Kimi K3. We dogfood it heavily at Databricks, and you’d be surprised how good it is.