DeepSeek’s V4 Flash model is listed on OpenRouter at 0.14 dollars per million input tokens and 0.28 dollars for output, with third-party platforms DeepInfra, GMICloud, and Baidu Qianfan quoting lower prices than the official API, including DeepInfra at roughly 64 percent of DeepSeek’s sticker price. The article discusses the possibility of DeepSeek raising its list price by 30x, which would still leave it the cheapest major model. An engineer from OpenCode argued the imminent price hike is a load-balancing tool, as server load has run at sustained peak since V4 Flash’s release, with demand exceeding capacity in a global 24-hour market. The key engineering factor is cache architecture: a 20,000-request experiment by researcher Su Chi Dan Dao showed DeepSeek maintaining 100 percent cache hit rates across 12-hour windows, while GLM collapsed to 25 percent after 5 minutes, with Kimi and GPT between the two. List price is not what most users pay, as stable prompts yield near-free inputs, while frequent context rewrites incur near-full price. A 30x hike would only affect the latter group. OpenCode’s advertised 10-dollar subscription does not always beat the official API, with real workload tests showing OpenCode Go running between 2x and 10x more expensive on long sessions due to cache misses. DeepSeek’s cheapest