New DeepSeek Flash model matches OpenAI's GPT-5.6 Luna at roughly 60 percent lower cost

Deepseek released V4 Flash 0731, an upgraded budget AI model.

It scored 50 on the Artificial Analysis Intelligence Index, up from 40, just one point below OpenAI’s GPT-5.6 Luna.

Thanks to a large cache discount and lower token usage, it costs about 60% less per task while keeping the same architecture and one-million-token context window.

The model also improved on GDPval and hallucination tests, narrowing the price-to-performance gap in AI models.