

DeepSeek V4-Flash and the Discount That Keeps Repeating
A frontier-class model, open weights, MIT license, and a price so low the cache hits cost a fraction of a cent per million tokens.
On July 31, 2026, DeepSeek released V4-Flash-0731 as a fully open-weight, MIT-licensed model on Hugging Face, the efficiency-tier sibling of its DeepSeek V4 family. The model carries 284 billion total parameters in a mixture-of-experts architecture, activating only 13 billion of them per token, and supports a 1 million token context window.
The pricing is what made the release notable again: around $0.14 per million input tokens and $0.28 per million output tokens at standard rates, with cache-hit pricing dropping to a fraction of a cent per million tokens. That is a small fraction of what comparable closed frontier models charge, and it arrives with full open weights rather than API-only access.
This is DeepSeek repeating a pattern it set with R1 in early 2025: ship a genuinely competitive model, license it permissively, and price it low enough that the economics of running a closed, proprietary alternative require active justification. Each repetition compresses the window in which a closed lab's pricing power goes unquestioned.
Enterprises that treat the cheapest open-weight release as a free substitute for their current model without re-running their own evaluation suite are making the same mistake every time this pattern repeats: a lower sticker price says nothing about reliability on your specific workload, only that it is now cheap enough to be worth testing seriously.
What it actually meant
Releasing a frontier-capable model at open weights and near-zero marginal cost keeps resetting what "state of the art" is allowed to cost, forcing every closed lab to actively justify its premium rather than assume it.
What enterprises get wrong today
Enterprises assume the cheapest open-weight model is a free lunch and skip the evaluation work, then discover the gaps in reliability and support that the sticker price never included.
Related capabilities
More from Models
Ask about this story
DeepSeek just did the open-weights price-crash thing again, this time with V4-Flash. Want to know the specs, how the pricing compares to closed models, or why DeepSeek keeps doing this?