DeepSeek’s Price Surge Fades into Obscurity? Flash's Peak Price Once Surpassed GPT-5.6 Luna

According to Dynamic Watch Beating monitoring, DeepSeek V4 has finally narrowed its most exaggerated price advantage. Starting on August 17, during the V4-Flash peak, the price per million tokens for input increased to $3, and for output increased to $9. The issue is that GPT-5.6 Luna, which is in the same category, was just discounted by OpenAI by 80%, with input now priced at only $0.20 and output at $1.20.

Based on the current exchange rate, Luna is priced at around $1.35 for input and $8.1 for output. In other words, during the DeepSeek Flash peak, not only did the price not become more competitive, but the input costs almost doubled, and the output is now slightly higher than Luna. The Artificial Analysis smart indexes for the two models are 50 and 51, with their performance almost neck and neck. Previously, DeepSeek used pricing to significantly outpace this tier of models, but now during the peak period, it can't even compete with Luna.

During off-peak hours, DeepSeek did manage to reclaim some cost-effectiveness. The Flash input price dropped to $1.5, similar to Luna; the output price of $4.5 is still about 44% cheaper. Even during heavy cache hits, DeepSeek's price advantage remains evident.

The V4-Pro is also entering a more competitive pricing range. During peak times, it is priced at approximately $1.33 for input and $4 for output, now very close to the low-priced flagship Meta Muse Spark, just slightly cheaper than Grok 4.6 at 2/6 dollars.

Therefore, after this round of price increases, DeepSeek is still relatively affordable, but being "affordable without a competitor" is now a thing of the past.