DeepSeek’s Flash update narrows the gap with OpenAI’s low-cost model

DeepSeek has updated its budget V4 Flash model, and The Decoder reports that the new 0731 version now sits close to OpenAI’s GPT-5.6 Luna on a widely watched benchmark while costing substantially less per task.

The model reportedly jumped 10 points to 50 on the Artificial Analysis Intelligence Index, placing it one point behind GPT-5.6 Luna. The Decoder says the DeepSeek model reaches that level at roughly 60 percent lower cost per task.

That matters because the market for everyday AI workloads is increasingly about price-performance, not only the most powerful flagship model. If a cheaper model is good enough for common tasks, developers can route more traffic to it and reserve expensive models for harder work.

Benchmarks are still only one signal. Real applications depend on latency, reliability, safety behavior, language coverage and tool use. But the update shows how quickly competition is moving in the lower-cost model tier.