DeepSeek
DeepSeek slashes prices with its new V4.1-Flash model
🧒 Discovery
A Chinese company just released a smart robot almost as skilled as the best ones, but far, far cheaper to run.
📰 Essential
Chinese company DeepSeek has launched V4.1-Flash, a model priced at just $0.003 per million cached input tokens off-peak, with results that sometimes beat Claude Opus 5 and GPT-5.6 on agentic tasks.
🔬 Expert
That $0.003 night rate applies only to cached-input hits; at peak hours and off cache, output pricing climbs to $0.60 per million tokens. On DeepSWE v1.1 the model edges out Claude Opus 5 and GPT-5.6 Sol, but lags well behind on Terminal-Bench 3.0. The competition is now increasingly about price-performance rather than raw intelligence.