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

Deepseek released a major update to its V4 Flash model, the "0731" version, that delivers comparable performance to OpenAI's GPT-5.6 Luna at significantly lower cost. The updated model jumped ten points on the Artificial Analysis Intelligence Index to reach a score of 50, placing it just one point behind OpenAI's offering while operating at approximately 60 percent lower cost per task.

This performance gain underscores the intensifying cost competition in the large language model space. Deepseek, the Chinese AI startup, has positioned its Flash variant as an efficiency-focused alternative to premium models. The gap closing between a budget-tier model and OpenAI's latest flagship suggests that raw capability differences are narrowing, even as price disparities remain substantial.

The timing matters. OpenAI's GPT-5.6 Luna represents the company's premium positioning in the market, targeting applications where performance justifies higher costs. Deepseek's ability to match it on benchmark scores at 60 percent lower pricing creates real pressure on procurement decisions for enterprises evaluating language model providers. Organizations running high-volume inference tasks stand to see meaningful savings by switching to Deepseek's updated model.

Benchmark scores, however, tell an incomplete story. Real-world performance varies across different task types. An index score of 50 versus 51 masks potential differences in reasoning, code generation, creative writing, or domain-specific applications. The Artificial Analysis Intelligence Index aggregates performance across multiple evaluations, but individual use cases may show different results.

The update reflects Deepseek's broader strategy of aggressive optimization. The company has consistently delivered capable models at lower cost than Western competitors, partly through engineering efficiency and partly through different training approaches. The Flash variant targets the volume segment, where marginal performance matters less than throughput and cost per token.

For the market,