DeepSeek's cheap model just beat its own flagship, so DeepSeek killed the flagship. Since September 14, every API call to V4 Pro gets answered by V4.1 Flash instead, at Flash prices, because Flash scores 74.2 on the DeepSWE coding benchmark against the flagship's 62.7. Flash costs 15 cents per million input tokens and 60 cents per million output off-peak. GPT-6 Astra and Claude Fable 5.1 both charge $10 and $50. On output that's a 40 to 80x gap depending on the hour, and for coding agents it…

boltdotnew@boltdotnew7h ago

Week one of open models on Bolt: builders picked the fastest one, by a landslide. Today: we're adding the one beta users say makes the best-looking sites. DeepSeek V4.1 Flash, now in Bolt Forge. 10x the usage of DeepSeek V4 Pro. https://t.co/7nOEN9LxMt https://t.co/UN6nWP7y74