What is DeepSeek V4 Flash?
DeepSeek V4 Flash is the versioned model ID for DeepSeek's V4 Flash line, deepseek/deepseek-v4-flash on Unifically. It first appeared in the V4 preview on April 24, 2026, and its final release, DeepSeek-V4-Flash-0731, followed on July 31, 2026. On September 10, 2026 DeepSeek retired that model and began answering the deepseek-v4-flash name with DeepSeek V4.1 Flash, billed at the Flash price. So today this ID gives you V4.1 Flash behavior: 1M tokens of context, up to 384K output tokens, and thinking on by default. It exists for clients and configs that already send the V4 Flash name.
Key features of DeepSeek V4 Flash
Same model as deepseek-flash today
Requests to this ID run DeepSeek V4.1 Flash, so answers, limits, and thinking behavior match `deepseek/deepseek-flash`. A pipeline built on the V4 Flash name keeps working without a code change, and picks up the newer model.
What V4 Flash 0731 was
The retired model was a 284B-parameter mixture-of-experts with 13B active parameters, the same architecture as the April preview with new post-training for agents. DeepSeek reported 82.7% on Terminal-Bench 2.1 and 54.4% on DeepSWE for it, and Artificial Analysis scored it 34 on its Intelligence Index.
What changed with V4.1 Flash
On DeepSeek's own table the model now behind this ID scores 90.6% on Terminal-Bench 2.1 and 74.2% on DeepSWE v1.1, where V4 Flash 0731 scored 82.7% and 54.4%. Artificial Analysis moves from 34 to 39 on its index, with output speed about the same at roughly 227 tokens per second.
Best for
Existing V4 Flash integrations
Clients, agents, and configs that already send the deepseek-v4-flash name.
Coding agents
The same agent performance as V4.1 Flash, since the same model answers.
Long-context work
Codebases and document sets read in one 1M token request.
Use cases
A coding agent or support bot set up for V4 Flash keeps running without a config change and now gets V4.1 Flash quality. The same goes for an OpenAI-compatible client or agent tool whose saved preset names deepseek-v4-flash. For anything new, start on deepseek/deepseek-flash so a future retirement of this name does not break your app.
Limitations
This ID no longer serves the V4 Flash 0731 model. If you need to reproduce old V4 Flash results exactly, the 0731 weights are public on Hugging Face under the MIT license, but the API will not return them.
DeepSeek describes the routing as temporary. Plan to move to deepseek/deepseek-flash.
Everything that limits V4.1 Flash applies here too: it trails Claude Opus 5 on the hardest Terminal-Bench suites, it writes long reasoning at high effort, and image input through Unifically is not confirmed yet. Peak-hour requests bill at 2x the off-peak rate.
When to use DeepSeek V4 Flash
Use this ID only to keep an existing V4 Flash integration running. For new work, call DeepSeek V4.1 Flash through deepseek/deepseek-flash, and use DeepSeek V4 Pro when the task depends on broad world knowledge.
API examples
Call DeepSeek V4 Flash by POSTing to /v1/chat/completions, the OpenAI-compatible endpoint shared by every language model on the platform. Parameter docs live at docs.unifically.com/models/llm/deepseek/deepseek-v4-flash.
curl -X POST https://api.unifically.com/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "deepseek/deepseek-v4-flash",
"messages": [
{ "role": "user", "content": "Write unit tests for this parser and list the edge cases you covered." }
]
}'
The response comes back synchronously with the answer in content and the thinking in reasoning_content. Set "stream": true to receive tokens as they generate.
FAQs
People also ask
deepseek/deepseek-v4-flash, called through the OpenAI-compatible POST /v1/chat/completions endpoint with your Unifically API key.
DeepSeek V4.1 Flash. DeepSeek retired the V4 Flash model on September 10, 2026 and routes the deepseek-v4-flash name to V4.1 Flash, billed at the Flash price. Output from this ID matches deepseek/deepseek-flash.
Use deepseek/deepseek-flash for new code. Keep deepseek/deepseek-v4-flash only if an existing config or client already sends it. DeepSeek calls the V4 Flash routing temporary, so this name may stop working later.
Yes. DeepSeek models bill at 2x the off-peak rate during peak hours, Monday to Friday from 01:00 to 04:00 and from 06:00 to 10:00 UTC. The rate is fixed when the request starts. Every other hour is off-peak, including all of Saturday and Sunday. This ID bills the same as deepseek/deepseek-flash; current rates are on the pricing page.
A 1M token context window and up to 384K output tokens in one response, the same limits as V4.1 Flash.
The final V4 Flash release, published on July 31, 2026. It kept the 284B-parameter preview architecture with 13B active parameters and was re-trained after pre-training for agent work. DeepSeek reported 82.7% on Terminal-Bench 2.1 and 54.4% on DeepSWE for it.
