DeepSeek-V4-Flash-High cracks top 7 in Frontend Code Arena, and it costs almost nothing to run

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

Deepseek V4 Flash scores 50 vs Luna's 51 at 60% lower cost per task (98% cache discount). For enterprise IT, cost-per-task now drives vendor choice as performance converges—budget arbitrage replaces feature differentiation.