DeepSeek V4 Flash 0731

DeepSeek released 2026-07-31

$0.105 per million tokens, balanced

Nothing is both better and cheaper.

Under a balanced workload, no other model in the catalogue scores higher and costs less. This model is on the value frontier.

Our take

editorial — not a measurement

Absurdly cheap for what it is: $0.44/$1.32 at peak, $0.22/$0.66 off-peak, with a 1.31M context window and 384k max output. The output ceiling is worth noting — 384k is triple what most frontier models allow, which makes it viable for bulk generation tasks that other models simply cannot complete in one pass. Same time-of-day billing and same aggressive cache discount as V4 Pro. If your workload is text-only and tolerant of a non-frontier model, this is close to the price floor for a capable long-context model.

Strengths

  • $0.22/$0.66 off-peak — near the price floor for this capability
  • 1.31M context and 384k max output
  • Cache hits at $0.007 off-peak
  • Open weights on Hugging Face

Weaknesses

  • Time-of-day pricing complicates budgeting
  • Text-only
  • Not frontier capability
  • Licence unverified

Reach for it when

  • Very large single-pass generation
  • Off-peak bulk processing
  • Cost-floor long-context work

Avoid it if

  • You need multimodal input
  • You need frontier reasoning

Sources: api-docs.deepseek.com

Every price dimension

Input$0.080/M
Output$0.180/M
Cached input$0.016/M

Independent scores

Intelligence51.8
Coding69.1
Agentic48.4

Sources are listed separately rather than averaged. Across the 62 models both have scored they correlate at r = 0.806 — close agreement overall, but four models rank very differently between them, and a blended score would hide exactly those.

Best rankings by task

  • uicomponent #30of 140 1266
  • 3d #32of 136 1256
  • website #33of 151 1262
  • gamedev #33of 144 1249

Capability

Context window1.3M
Max output384K
Input modestext
Tool useyes
Reasoningoptional

Provenance

Price sourceopenrouter.ai
Fetched2026-08-24
Quality dataverified

DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows....

Markdown for LLMs