Open weight, MIT

DeepSeek V4 online, pay per use

DeepSeek V4 is the open-weight mixture-of-experts family that made DeepSeek the single largest source of tokens on OpenRouter. It comes in two sizes, V4-Pro for the hardest reasoning and coding, and V4-Flash for fast, high volume work, both with a 1 million token context. On avots.ai you run them pay per use, no subscription, and the same balance also covers Claude Opus 5, GPT-5.6, Gemini and 40+ image, video and music models.

Use DeepSeek V4 now ✦ See pricing
At a glance

DeepSeek V4 specs

Open weight MoE

Mixture-of-experts under the MIT license. V4-Pro has about 1.6T total parameters with 49B active, V4-Flash has 284B total with 13B active.

1M token context

Both variants default to a 1 million token context window with up to 384k output, enough for whole repositories and long documents in one pass.

Sparse attention

DeepSeek Sparse Attention with token-wise compression. At 1M context V4-Pro uses about 27 percent of the inference compute and 10 percent of the KV cache of V3.2.

Top coding scores

V4-Pro at max reasoning hits a LiveCodeBench Pass@1 of 93.5 and about 80.6 percent on SWE-bench Verified.

Very low price

V4-Pro runs at about 0.44 USD input and 0.87 USD output per million tokens, V4-Flash at about 0.14 and 0.28.

Vision optional

V4 focuses on text and code reasoning. For image input pair it with Gemini or GPT-5.6 on the same avots balance.

Real numbers

DeepSeek V4-Pro vs the frontier

Coding benchmarkDeepSeek V4-ProGemini 3.1 ProClaude Opus 4.6 Max
LiveCodeBench Pass@1 (max effort)93.591.788.8
SWE-bench Verifiedabout 80.6%stronghigh
Context window1M tokens1M tokens200K tokens
Weightsopen, MITclosedclosed
Price per 1M in / out$0.44 / $0.87$2 / $12higher

On DeepSeek's own coding evaluation V4-Pro at max reasoning tops LiveCodeBench, ahead of Gemini 3.1 Pro and Claude Opus 4.6 Max, at a fraction of the price and with open weights. Numbers are from DeepSeek's release notes and independent write-ups.

Three steps

How to use DeepSeek V4

Create an account

Sign up on app.avots.ai with email, Google or Telegram. Takes a minute.

Top up from 0.99 EUR

1,000 tokens cost 0.99 EUR. No plan to choose, no renewal, the balance sits there until you use it.

Pick V4-Pro or V4-Flash

Choose a DeepSeek V4 model in the picker, or let Agent Avots route for you. Every reply shows its token cost.

Sign up and start
One balance

Where you can use DeepSeek V4

Web app

app.avots.ai on desktop and mobile, no install needed.

Telegram

DeepSeek V4 in @AvotsAIbot, same balance, from any phone.

MCP

Use it inside Claude Desktop, Cursor or Cline via the MCP server.

Try DeepSeek V4 for under a euro

1,000 tokens cost 0.99 EUR and DeepSeek V4 replies cost a fraction of a cent. The same balance unlocks Claude Opus 5, GPT-5.6, Gemini, Kimi K3, Veo video and Nano Banana images.

Sign up and chat ✦
Guides

Related pages

GLM-5.2

Z.ai's open-weight coding model, 1M context, MIT license. Live now.

Kimi K3

Moonshot's frontier open model, number one at coding, 1M context.

GPT-5.6

OpenAI's newest Sol, Terra and Luna tiers, 1M context. Live now.

avots.ai runs 40+ AI models for chat, image, video and audio on one balance. See the full list on the Tools page or open the app and start creating.

Browse all model pages Open the app
FAQ

DeepSeek V4 questions

What is DeepSeek V4?
DeepSeek V4 is an open-weight mixture-of-experts model family from DeepSeek, released under the MIT license. It ships in two sizes: V4-Pro with about 1.6 trillion total parameters and 49 billion active per token, and V4-Flash with 284 billion total and 13 billion active. Both default to a 1 million token context window with up to 384k output.
How much does DeepSeek V4 cost on avots.ai?
Pay per use, no subscription. Per million tokens V4-Pro is about 0.44 USD input and 0.87 USD output, and V4-Flash is about 0.14 and 0.28, so a chat reply costs a fraction of a cent. Top up from 0.99 EUR and every reply shows its token cost before you run.
Is DeepSeek V4 good at coding?
Yes, coding is its strongest area. At max reasoning effort V4-Pro reaches a LiveCodeBench Pass@1 of 93.5, the highest score among models DeepSeek evaluated, ahead of Gemini 3.1 Pro at 91.7 and Claude Opus 4.6 Max at 88.8. It also scores about 80.6 percent on SWE-bench Verified.
Is it the real DeepSeek V4?
Yes. avots.ai routes to the same DeepSeek V4-Pro and V4-Flash models through official APIs. Same weights, same quality, pay per use instead of a monthly plan.
Can I use DeepSeek V4 with Claude and GPT on one balance?
Yes. One balance covers 40+ models: DeepSeek V4, Claude Opus 5, GPT-5.6, Gemini and Kimi K3 for chat, plus Veo, Sora and Nano Banana for media. Use them in the web app, Telegram, MCP or the API.
Try it on avots ✦
Start creating