On our own GPUs
DeepSeek V4 Pro API
1.6 trillion parameters, thinking and non-thinking modes, MIT licensed.
Context
1M
Max output
384K
Serving precision
BF16
Licence
MIT
DeepSeek V4 Pro is DeepSeek’s open-weight flagship, released on 24 April 2026. A 1.6T-parameter mixture-of-experts with 49B active per token, a 1-million-token context window and up to 384K output tokens. Like its sibling V4 Flash it ships dual thinking / non-thinking modes, and both are released under the MIT licence. It is currently the strongest reasoning tier available in the open-weight camp — the model to hand your hardest work to.
Open weights, deployed in facilities we build out and lease directly, operated by us. We resell nobody else’s account quota.
Specifications
- Model ID
- deepseek-v4-pro
- Published by
- DeepSeek
- How it is served
- On our own GPUs
- Licence
- MIT
- Total parameters
- 1.6T MoE
- Active parameters
- 49B
- Context
- 1M
- Max output
- 384K
- Serving precision
- BF16
- Input modalities
- Text
- Released
- 2026-04-24
- Price
- See pricing
What it is good at
- Reasoning only when the task earns it
- The same model answers fast in non-thinking mode and runs long chains when you switch modes on. You pay for depth on the requests that need it instead of on all of them.
- 384K tokens of output
- Combined with the 1M window, read-a-book-write-a-book workloads finish in a single call — no manual segmentation, no state to carry between requests.
- The strongest open reasoning under MIT
- This capability tier under an essentially unrestricted licence is one of the best deals in self-hosting. We run it on our own cluster with the weight version and precision published.
Where it fits
- Mathematics, algorithms and hard multi-step logic
- Defect localisation and cross-file refactors in large codebases
- Reports, translations and generation that need very long output
- The escalation tier behind a cheaper default model
Three steps
Change one line of base_url, put the model ID below in the `model` field, and leave the rest of your code alone.
model: "deepseek-v4-pro"
curl https://api.nexaix.net/v1/chat/completions \
-H "Authorization: Bearer $NEXAIX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-pro",
"stream": true,
"messages": [
{ "role": "user", "content": "Summarise this contract." }
]
}'import os
from openai import OpenAI
client = OpenAI(
# the only new line
base_url="https://api.nexaix.net/v1",
api_key=os.environ["NEXAIX_API_KEY"],
)
# everything below is your existing code
stream = client.chat.completions.create(
model="deepseek-v4-pro",
stream=True,
messages=[
{"role": "user", "content": "Summarise this contract."}
],
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.nexaix.net/v1", // the only new line
apiKey: process.env.NEXAIX_API_KEY,
});
// everything below is your existing code
const stream = await client.chat.completions.create({
model: "deepseek-v4-pro",
stream: true,
messages: [
{ role: "user", content: "Summarise this contract." }
],
});
for await (const chunk of stream) {
process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}About this model
V4 Pro or V4 Flash?
Pro is 1.6T with 49B active — the top reasoning tier, for hard work and escalation. Flash is 284B with 13B active at a much lower unit cost, for bulk production traffic. The usual pattern is Flash by default with Pro behind a difficulty check you control.
Does thinking mode cost extra?
Thinking mode emits chain-of-thought tokens, which are billed as output. See the pricing page for rates. Turn it on only where long-chain reasoning actually pays for itself.
Does DeepSeek V4 accept images?
No. Both shipped variants are text-only mixture-of-experts language models with no vision or audio capability. For multimodal work use Kimi K3 or MiniMax M3.
Also worth a look
- On our own GPUsKimi K32.8T mixture-of-experts, a 1M-token window, and vision built into the base.kimi-k3
- On our own GPUsGLM-5.2750B sparse MoE, a 1M window, MIT licensed — tuned hard for software engineering.glm-5.2
- On our own GPUsDeepSeek V4 Flash284B total, 13B active per token — the lowest unit cost in the V4 generation.deepseek-v4-flash
Run your own eval on it
Take the trial credit, run the evals you already have against this model, and then decide whether it gets your production traffic.
