DeepSeek-V4.1-Flash
DeepSeek · Undisclosed (MoE, ~8B active)
The nail that sells document skill for ¥0.6
DeepSeek’s first low-cost general-purpose native multimodal model, positioned very differently from the layout-only DeepSeek-OCR 2. Composite 87.9 with documents 95 and charts 92 — near-flagship document skill at ¥0.6/1M output tokens, dominating five models on this board that cost more than ten times as much.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- DeepSeek-V4.1-Flash87.9
Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.
Public benchmarks
as reported, not converted| Benchmark / metric | Value | Note |
|---|---|---|
| OmniDocBench v1.5 | 0.104 | Edit distance, near flagship |
| 输出单价 | ¥0.6 | Lowest generalist price here |
| ChartQA | 88.1 | Official claim |
Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.
Scenario mapping
five stress tests, one scale| Scenario | Score | Observation |
|---|---|---|
| Chinese receipt OCR | 95 · Handwriting weaker; print excellent | |
| Chart data extraction | 92 · Far ahead of its price class | |
| 30-min video QA | 74 · Coarse summaries only | |
| Mobile GUI control | 82 · Usable coordinates, average planning | |
| Chinese cultural context | 86 · Factual; weak on everyday context |
Strengths and limitations
Strengths
- Documents 95 at ¥0.6/1M output tokens — the outermost point on our price-performance frontier.
- MIT open weights with zero commercial strings — the most permissive licence among the four new models.
- Short reasoning traces and low time-to-first-token give visibly better throughput on batch document jobs.
Limitations
- Video timing at 78 is among the lowest here; long video supports only coarse summaries.
- Generation and editing at 72 — effectively no usable text-to-image or Chinese glyph rendering.
- 128K context is small by 2026 standards; whole manuals still need chunking.
Price and real cost
List price
¥0.6/M tokens
≈ ¥0.6 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥0.4 (assuming input at 25% of output price).
Use your volumeAbout DeepSeek-V4.1-Flash
Is DeepSeek-V4.1-Flash open source?
Licence: Open weights (MIT). Weights are available, but check the exact terms and filing requirements before commercial use.
How much does DeepSeek-V4.1-Flash cost?
≈ ¥0.6 / 1M output tokens. It ranks #12 with a composite of 87.9. Real bills depend on resolution, tiling, caching and volume discounts.
What is DeepSeek-V4.1-Flash best for?
High-volume receipt, contract and scan structuring; Cost baseline and control group for document pipelines; Commercial on-prem work that needs an MIT licence。
Why does DeepSeek-V4.1-Flash score 87.9?
The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Document parsing & Chinese OCR; its weakest is Generation & editing.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.