DeepSeek’s New Model Nearly Matches GPT-6 Astra on Design, at 1.4% of the Cost

DeepSeek released a new model that scores close to GPT-6 Astra on a design-focused benchmark, while running at a fraction of the cost. The reported figure is 1.4% of GPT-6 Astra’s cost for comparable design performance.

Design benchmark results

On the design evaluation, DeepSeek’s latest model posted results that “nearly match” GPT-6 Astra. The comparison is framed around design capability rather than general text performance.

Cost gap: 1.4% versus GPT-6 Astra

The article’s central efficiency claim is cost. DeepSeek’s model is described as reaching similar design outcomes at 1.4% of the cost attributed to GPT-6 Astra.

What the comparison is measuring

The benchmark comparison focuses on design tasks, using GPT-6 Astra as the reference point. The cost figure is presented as the main differentiator alongside performance.

Published timing

The report was published on 2026-09-10T21:16:03.

Why it matters

For model buyers and operators, the combination of near-competitive benchmark performance and a sharply lower cost per run changes the economics of deploying design-capable AI. It also adds a new data point to how quickly frontier performance is being compressed into more affordable systems.