Skip to content

Ask AI

I answer questions about Radeon Cloud from these docs and link to the pages I used.

Answers are generated from the docs by a model on Token Factory and can be wrong; the docs are authoritative. Don't enter API keys or other secrets here.

DeepSeek-V4-Flash-Vision-Exp

POST /v1/chat/completions model: DeepSeek-V4-Flash-Vision-Exp

A vision tower bolted onto the DeepSeek-V4-Flash weights; the Exp in the name is experimental. It has a one-million-token context and takes images, like DeepSeek-V4.1-Flash and MiMo-V2.6-Flash.

The language side shares its architecture with the text model, field for field. The table below is read from the config.json of the weights this endpoint actually loads.

ArchitectureDeepseekV4ForCausalLM, model_type = deepseek_v4
Layers43
Hidden size4,096
Attention64 query heads, 1 KV head, head dim 512 (MLA)
Mixture of experts256 routed + 1 shared, 6 activated per token, expert intermediate 2,048
Sparse indexa DSA indexer with index_n_heads / index_topk
Sliding window128
Vocabulary129,280
Native context1,048,576
QuantisationFP8 e4m3, block size [128, 128], dynamic activation scaling, ue8m0 scale format

The vision tower (same config.json, expressed as flat vision_* keys rather than a nested vision_config):

Layers / width / heads32 layers, 1,024 wide, 16 heads
Patch size14
Intermediate size2,816
Downsample ratio3
Per-image token ceiling384
Minimum pixels147,456
Maximum width:height ratio8

vision_max_n_token = 384 is a hard ceiling: a larger image is compressed to 384 tokens, so very high-resolution inputs do not cost linearly more — and do not carry more detail either.

Context length1,048,576 tokens
Input modalitiestext + images
Streaming✅
Tool calling✅
JSON output✅ json_object
Thinking✅ — off by default, ask for it explicitly
Stabilityexperimental
How to sendan {"type":"image_url","image_url":{"url":"data:image/png;base64,…"}} part in content
Meteringusage.prompt_tokens_details.image_tokens
Per-image ceiling384 tokens (above)
Terminal window
curl https://developer.amd.com.cn/radeon/api/v1/chat/completions \
-H "Authorization: Bearer $RADEON_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "DeepSeek-V4-Flash-Vision-Exp",
"messages": [{
"role": "user",
"content": [
{ "type": "image_url", "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." } },
{ "type": "text", "text": "What does this image say?" }
]
}]
}'

As permissive as the text model: a system message may sit at any position, there may be more than one, and the role may be spelled either system or developer.

This differs from Qwen3.8-27B, which accepts at most one system, and only at the start. For one client driving both, follow the stricter shape.

Off by default. With reasoning_effort omitted, reasoning_tokens comes back as 0; you have to ask for thinking explicitly.

All six reasoning_effort values are accepted (minimal, low, medium, high, xhigh, max, plus none and omission) — the most permissive model on this endpoint.

Thinking textchoices[0].message.reasoning
Token countusage.completion_tokens_details.reasoning_tokens
Top-level aliasusage.reasoning_tokens — this model does emit it
Context window1,048,576 tokens, counted as prompt plus output
JSON outputresponse_format: {"type": "json_object"}
Turning thinking onreasoning_effort (or the equivalent reasoning.effort)