DeepSeek-V4-Flash-Vision-Exp
We are excited to introduce DeepSeek-V4-Flash-Vision-Exp, our first experimental multimodal model in the DeepSeek-V4 family. It builds on the DeepSeek-V4-Flash architecture by incorporating visual modules and undergoing continued training to unlock visual understanding capabilities.
Params
304.6 B
Context
1,048,576
Downloads 30d
836 K
Likes
932
Download history
daily snapshots · 31 days
▲ 1 K in the last 30 days (0.2%)
989 K836 K
Sep 24Oct 4Oct 14Oct 23
989 K836 K
Sep 23Oct 3Oct 13Oct 23
Can you run it?
Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.
| File | Quant | Size | Est. VRAM | Verdict on RTX 4090 · 24 GB |
|---|---|---|---|---|
| model.safetensors | i8 | 167.8 GB | 230.8 GB | ❌ Won’t fit |
| model.safetensors (bf16, full) | bf16 + 1024K ctx | 167.8 GB | 6,034.3 GB | ❌ Won’t fit |
Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.
Use via API
via OpenRouter · USD per 1M tokens · synced 2026-10-23$0.22
input / 1M
$0.65
output / 1M
$0.0069
cached input / 1M
$0.86
1M in + 1M out
- Context window
- 1,048,576 tokens
- Max output
- 262,144 tokens
- Providers
- 3
- Cheapest provider
- DeepInfra · $0.22 / $0.65
- Best throughput
- —
- AA Intelligence / Coding / Agentic
- 34.8 / 65 / 47.5
✓ Tool calling
✓ Reasoning
✓ Structured outputs
✓ Image input
✗ File input
✗ Audio input
✓ Prompt caching
Specifications
- Architecture
- DeepseekV4ForCausalLM
- Parameters
- 304.6 B
- Tensor type
- I8
- Context length
- 1,048,576
- Vocabulary
- 129,280
- Layers / heads
- 43 / 64
- Licence
- mit
- First seen on the Hub
- 2026-08-31
- Training datasets
- undisclosed
- Added to our catalog
- 2026-09-23
Compare with any image-text-to-text model