DeepSeek V4 Pro Costs 15x More

The launch reads like a tradeoff, not a clean win: DeepSeek bought more capability and longer-context efficiency, but max-effort usage now looks like a premium model problem.

• –DeepSeek’s own technical report says V4 Pro needs only 27% of the single-token inference FLOPs and 10% of the KV cache versus V3.2 at 1M context, so the architecture is genuinely more efficient.
• –Artificial Analysis says evaluating V4 Pro on its Intelligence Index cost $1,071.28 and produced 190M output tokens, which explains the “15x” sticker shock and the verbosity complaint.
• –The model scores 52 on the AA Intelligence Index, but the comparison page still shows Gemini 3.1 Pro Preview at 57, so the headline is more about cost and trajectory than a clean benchmark takeover.
• –For developers, the practical takeaway is that V4 Pro looks like the flagship to use when raw capability matters, while V4 Flash is likely the better default when spend and throughput matter.
• –This is a stronger model line than V3.2, but it no longer fits DeepSeek’s old “near-frontier at bargain pricing” story in Pro mode.