Where Is DeepSeek V4 API Available? Official, Model IDs, and Platform Support

If you’re searching for “DeepSeek V4 API”, the first thing to know is this: DeepSeek does provide a V4 API officially, but the actual official model IDs are not deepseek-v4. They are deepseek-v4-pro and deepseek-v4-flash.

That naming mismatch is where most confusion starts. Websites, news posts, and third-party model pages often say “DeepSeek V4”, but what really matters in production is the exact model ID, whether the platform is OpenAI-compatible, and whether the model is actually available today.

This guide summarizes the public official information available on 2026-04-29 and answers two questions:

1. Which platforms actually offer DeepSeek V4 API access?
2. What is the fastest way to make your first request?

---

Quick Answer

• The official DeepSeek platform is available and remains the safest default choice
• DeepSeek recommends OpenAI-compatible access with base_url set to https://api.deepseek.com
• The official V4 models to use are:
  • deepseek-v4-pro
  • deepseek-v4-flash
• Legacy aliases deepseek-chat and deepseek-reasoner still appear in the docs, but DeepSeek marks them for deprecation on 2026-07-24
• As of 2026-04-29, the platforms where V4 can be clearly verified from public official pages include:
  • DeepSeek official
  • Alibaba Cloud Bailian
  • Together AI
  • Fireworks AI
  • Novita AI
• OpenRouter has a deepseek/deepseek-v4 page, but that page explicitly says the model is not available
• Volcengine Ark, Tencent Cloud, and Replicate do not have enough public official evidence here to safely describe as “DeepSeek V4 supported”

---

DeepSeek V4 API Platform Summary

Platform	Available on 2026-04-29	Verified model name / ID	OpenAI-compatible	Notes
DeepSeek official	✅ Yes	deepseek-v4-pro, deepseek-v4-flash	✅	Best documentation, canonical naming
Alibaba Cloud Bailian	✅ Yes	deepseek-v4-pro, deepseek-v4-flash	Platform-specific integration path	Good fit for mainland China deployments
Together AI	✅ Yes	deepseek-ai/DeepSeek-V4-Pro, deepseek-ai/DeepSeek-V4-Flash	✅	Easy if you already use OpenAI SDKs
Fireworks AI	✅ Yes	deepseek-v4-pro	✅	Official docs support OpenAI client usage
Novita AI	✅ Yes	Deepseek V4 Pro, Deepseek V4 Flash	✅	Public model pages show pricing and context
OpenRouter	⚠️ Page exists, unavailable that day	deepseek/deepseek-v4	✅	Public page explicitly says not available
SiliconFlow	⚠️ DeepSeek models exist, but no public V4 entry confirmed here	DeepSeek-V3.1, DeepSeek-V3.2, etc.	✅	Do not overstate as confirmed V4 support
Volcengine Ark	❓ Not confirmed from public official pages here	—	—	Avoid claiming support without stronger evidence
Tencent Cloud	❓ Not confirmed from public official pages here	—	—	Public docs found here focus on Hunyuan
Replicate	❓ No official DeepSeek V4 entry confirmed here	—	Mostly its own prediction API style	Avoid claiming official V4 support

Methodology: only platforms that can be verified from public official docs or official model pages are marked as confirmed.

---

How to Use the Official DeepSeek V4 API

If you just want the fastest path to a working request, start with the official API.

Step 1: Create an API key

Create a DeepSeek API key, then export it locally:

export DEEPSEEK_API_KEY="your DeepSeek API key"

Step 2: Use the OpenAI-compatible endpoint

Official base URL:

https://api.deepseek.com

Most common endpoint:

https://api.deepseek.com/chat/completions

Step 3: Send your first request

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
      {"role": "user", "content": "Explain DeepSeek V4 API in one sentence."}
    ],
    "stream": false
  }'

If you care more about speed and cost, switch the model to deepseek-v4-flash.

---

Using the OpenAI SDK with DeepSeek

If your stack already uses the openai SDK, migration is simple: change the base_url, api_key, and model name.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["DEEPSEEK_API_KEY"],
    base_url="https://api.deepseek.com",
)

resp = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[
        {"role": "user", "content": "What is a cache hit in LLM APIs?"}
    ],
)

print(resp.choices[0].message.content)

That is also why third-party providers emphasize OpenAI compatibility: you often do not need to change your application logic.

---

How to Choose a Platform

1. DeepSeek official

Best for developers who want the clearest naming, the latest official rollout, and the least ambiguity.

Why it is the safest default:

• Canonical model names
• Best official docs
• Clear OpenAI-compatible entry point
• Explicit migration guidance for legacy aliases

Recommended choices:

• deepseek-v4-pro for higher-quality, heavier tasks
• deepseek-v4-flash for speed and lower cost

2. Alibaba Cloud Bailian

Best for teams in China, or organizations already standardized on Alibaba Cloud.

Publicly verifiable details:

• Official model pages list deepseek-v4-pro and deepseek-v4-flash
• Mainland China and global deployment scopes differ
• Mainland listing shows China North 2 (Beijing)
• Global listing shows US (Virginia)

For domestic production and procurement workflows, this is often easier than overseas providers.

3. Together / Fireworks / Novita

Best for teams already building on model aggregators and unified OpenAI-compatible gateways.

Shared advantages:

• Public official pages clearly list V4 models
• OpenAI-compatible integration is straightforward
• Good fit if you already route multiple providers through one abstraction layer

The main differences are in:

• Model ID naming
• Pricing
• Cache billing
• Context window and output limits
• Whether both Pro and Flash are available at the same time

4. OpenRouter

Do not assume support just because a model page exists.

As of 2026-04-29, OpenRouter shows a deepseek/deepseek-v4 page, but the page explicitly says:

The model “deepseek/deepseek-v4” is not available

So the safe rule is: a model page is not the same thing as an actually callable model.

---

The Naming Trap: DeepSeek V4 vs deepseek-v4-pro vs deepseek-chat

This is the single most important clarification.

1. DeepSeek-V4 is the product or series name

That is the label you usually see on websites, launch posts, and model catalogs.

2. The actual official API model IDs are:

• deepseek-v4-pro
• deepseek-v4-flash

3. Legacy aliases still exist, but should not be your default for new projects

The official docs still mention:

• deepseek-chat
• deepseek-reasoner

They are compatibility aliases and already marked for deprecation. New integrations should use the V4 model IDs directly.

---

What to Check Before You Ship

Before you use DeepSeek V4 in production, do not stop at “this platform has it.” Check these four things:

1. Input and output pricing
2. Cache hit vs cache miss billing
3. Context window size
4. Max output token limit

DeepSeek’s official pricing page is the best place to understand the baseline. Third-party platforms may change model IDs, pricing logic, or caching behavior.

The safest workflow is:

• understand the model from DeepSeek’s own docs first
• verify the final platform-specific model page before launch

---

Four Common Pitfalls on Third-Party Platforms

1. Model names may differ completely

For example:

• official: deepseek-v4-pro
• Together: deepseek-ai/DeepSeek-V4-Pro
• OpenRouter: deepseek/deepseek-v4

Never copy a model name from one provider to another blindly.

2. OpenAI-compatible does not mean perfectly identical

Providers may differ in:

• supported parameters
• reasoning or thinking modes
• function calling
• JSON mode
• streaming behavior

So “it runs” does not always mean “it behaves exactly the same.”

3. Domestic cloud platforms may add account, region, or compliance requirements

Alibaba Cloud Bailian already shows region and deployment-scope differences. In enterprise environments, you should also confirm:

• whether real-name verification is required
• whether an enterprise account is needed
• whether you must pick a supported region
• whether additional service activation is required

4. Pricing changes quickly

Model pricing moves fast. Any price in an article should be treated as a snapshot for that day, not a guarantee.

---

If You Want the Safest Default Choice

My recommendation is simple:

• Solo developer / fastest path: use the official DeepSeek API first
• China-based company or team: check Alibaba Cloud Bailian first
• Already on global model gateways: compare Together, Fireworks, and Novita
• Seeing an OpenRouter model page: confirm actual availability before integrating

---

Summary

As of 2026-04-29, DeepSeek V4 API is available on more than just the official platform. But when you actually integrate it, the important questions are not just “does this site mention DeepSeek V4?” They are:

1. What is the real callable model ID on this platform?
2. Is the model actually available today?
3. Does the platform fit your existing OpenAI SDK or gateway workflow?

If you want the least risky route, start with the official DeepSeek API and use deepseek-v4-pro or deepseek-v4-flash directly. Move to Bailian, Together, Fireworks, or other platforms only when you specifically need multi-provider routing, cloud procurement alignment, or different cost controls.

References:

• DeepSeek API Docs
• DeepSeek Pricing
• Alibaba Cloud Bailian Models
• OpenRouter DeepSeek V4 page
• Together Models
• Fireworks Models
• Novita Models