DeepSeek V4 Flash 0731 Input Price Jumps 1308% — Output Gets Cheaper

What Changed

DeepSeek quietly repriced V4 Flash 0731 on its API. Input costs moved from $0.0071 to $0.10 per 1M tokens — a +1308% increase. Output costs moved in the opposite direction, falling from $1.28 to $0.30 per 1M tokens.

That's a genuinely unusual pricing structure shift: the model got dramatically more expensive to prompt and meaningfully cheaper to run to completion.

Who This Hurts, Who It Doesn't

The impact depends almost entirely on your input-to-output ratio.

High-input workloads — long-context retrieval, document summarization with large source material, or RAG pipelines stuffing big chunks into every prompt — will feel this hard. A workload burning 10M input tokens daily just went from roughly $0.07 to $1.00 in input costs alone.

High-output workloads — code generation, long-form drafting, or anything where the model writes substantially more than it reads — actually come out ahead. Output dropping from $1.28 to $0.30 per 1M tokens is a significant saving if that's where your spend sits.

Balanced workloads will need to do the math. Run your last 30 days of token logs through both price points before assuming this is a net negative.

What to Do Now

Check your input/output split first. If input tokens dominate, this repricing warrants a serious look at alternatives. If output tokens dominate, you may be better off than before.

For full current specs and an ongoing price history, see DeepSeek V4 Flash 0731 — live specs & price history.

Pricing changes this asymmetric are worth watching — they often signal shifts in how a provider wants the model used.

Ezra, Scout AI Team