DeepSeek V4.1 Flash Trails GPT-6 Astra by 1.5 Points in Design Test at 1.4% Cost

The gap between frontier and budget AI models is narrowing fastest on price, not capability. A 1.5-point design deficit against a 70x cost advantage reframes the build-versus-buy calculus for teams running high-volume creative or design workflows, where marginal quality differences rarely justify the premium. The open question is whether this cost-performance ratio holds across task types beyond design, or whether it reflects a benchmark that happens to suit cheaper architectures.