DeepSeek V4 Flash Vision Exp
DeepSeek · released Aug 31, 2026 · deepseek-ai/DeepSeek-V4-Flash-Vision-Exp
- Type
- Open weightsMIT License
- Params
- 305B
- Context
- 1M
37B active per word · about 786K words of context
Our take
Written Sep 30, 2026DeepSeek V4 Flash Vision Exp is a specialist you can download and run yourself: it scores well on mathematics, data analysis and agentic coding, but sits near the bottom of the field on plain code generation. The licence allows commercial use, changes and redistribution (MIT).
Pick it for mathematical and data-analysis work, or for agentic coding run inside a harness, where it ranks 5th of 58 on LiveBench Agentic Coding as of 25 Jun 2026. It takes images alongside text, so a diagram or screenshot can go straight into the request. Skip it if you need reliable plain code generation, if you want a model that fits comfortably on your own machine, or if you need measured quality outside the LiveBench categories listed here.
The case for it
- 87.81% on LiveBench Mathematics and 79.48% on LiveBench Data Analysis, so it is a reasonable first trial for maths and table work.
- 65.1% on LiveBench Agentic Coding inside an agent harness, 5th of 58 as of 25 Jun 2026, which is the job to start it on.
- Text and image input with written output, so a screenshot or diagram does not have to be described in words first.
- The licence allows commercial use, changes and redistribution (MIT).
The case against it
- 68.2% on LiveBench Coding, 56th of 58 as of 25 Jun 2026, so it is among the weakest measured options for writing code from scratch.
- 305 billion parameters in total, 37 billion active per token, so running it yourself needs substantial memory rather than a workstation.
- Its measured results sit in the LiveBench categories listed here; nothing here covers other benchmarks or tasks.
How good is it?
EverydayGeneral questions and everyday reasoning
Not yet scored on Arena Text (overall). It is on LiveBench Data Analysis, in 9th of 58 with 79.48.
CodingWriting and fixing code on its own
Not yet scored on Arena Coding. It is on LiveBench Coding, in 56th of 58 with 68.2.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on LiveBench Agentic Coding, in 5th of 58 with 65.1.
WritingDrafting and rewriting prose
Not yet scored on Arena Creative Writing. It is on LiveBench Language, in 24th of 58 with 80.36.
Boards this model appears on that none of the ratings above are built on.
Every published score for this model8 scoresEvery figure we hold, from 8 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M3 Ultra (80-core GPU) · 512 GB
Room to spare. 184.8 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked 4 hours ago — each listing carries its own date.
Some hosts sell this model at two prices: on their own price list (“direct”) and on their OpenRouter listing (“through OpenRouter”). Where the two differ, the row shows both, each with the date we last read it.
- per 1M tokens
- $0.22 in / $0.65 out
- Context served
- 1M
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.22 / $0.65checked 4 hours ago | 1M | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp8Direct and through OpenRouter | $0.44 / $1.32directchecked 4 hours ago$0.22 / $0.65through OpenRouterchecked 4 hours ago | 1M262K max reply through OpenRouter | 70 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Novita AIDirect and through OpenRouter | $0.44 / $1.32checked 4 hours ago | 1M393K max reply through OpenRouter | 85 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| GMICloudfp8Through OpenRouter | $0.44 / $1.32checked 4 hours ago | 1M944K max reply | 80 tok/s | No | Yesunknown period | Unknown |
| SiliconFlowfp8Through OpenRouter | $0.44 / $1.32checked 4 hours ago | 1M393K max reply | 91 tok/s | No | No | Confirmed |
DeepInfra
Novita AIAcross the 5 listings we hold: 4 say they do not train on prompts (2 of them only through OpenRouter), 0 say they do and 1 does not say. 3 appear in the zero-retention registry we check (2 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| DeepInfrafp8Direct and through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIDirect and through OpenRouter | ✓ | ✓ | ✗ |
| GMICloudfp8Through OpenRouter | ✓ | ✓ | ✗ |
| SiliconFlowfp8Through OpenRouter | ✓ | ✗ | ✗ |
Tool calling: 5 of 5 listings say yes. JSON output: 4 of 5 listings say yes, 1 says no. Strict schema: 2 of 5 listings say yes, 3 say no.
When we formed this view
Recent changes
What moved
cache read −90% ($0.069 → $0.007 per 1M tokens)What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 5 listings does not say whether it trains on prompts, and 2 answer only through OpenRouter, not for their own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsMIT License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
MIT License
Fully permissive: do anything with attribution. No patent grant, unlike Apache-2.0.
Identifiers
- Hugging Face
- deepseek-ai/DeepSeek-V4-Flash-Vision-Exp
- Architecture
- Mixture of experts
- Takes in, gives back
- Text and images in, text out
- Catalogue slug
- deepseek-deepseek-v4-flash-vision-exp