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

Editor's take

Deepseek's V4 Flash LLM, post-update, has achieved a score of 50 on the Artificial Analysis Intelligence Index, placing it within a single point of OpenAI's GPT-5.6 Luna while operating at a substantially lower cost.

This development is significant for the broader LLM market, particularly for developers and businesses seeking more economical yet performant models. Deepseek's ability to approach the capabilities of a leading proprietary model like GPT-5.6 Luna at a 40% premium suggests increasing viability for open-source or more cost-conscious alternatives in enterprise applications, potentially challenging the established dominance of companies like OpenAI and Anthropic, whose models often command higher inference costs.

Future developments to watch include whether Deepseek can sustain this performance trajectory and further close the gap on key benchmarks, and how OpenAI and others will respond to this cost-efficiency challenge. The long-term impact will depend on the scalability and reliability of Deepseek's offering in real-world deployments, and whether its performance holds across a wider range of complex tasks, not just those measured by the Artificial Analysis Intelligence Index.