DeepSeek V4 Flash API

Playground

DeepSeek V4 Flash

Free public preview

ClearControls

Start a conversation with DeepSeek V4 Flash

Send a message to test tone, reasoning, and instruction following in real time.

Press Enter to send and Shift + Enter for a new line.

Send

Endpoint support

DeepSeek V4 Flash capabilities on Geodd

These values describe Geodd's published endpoint capabilities, not capabilities inherited from the base model. Features listed as supported show Yes; otherwise they show No. Independent test results are not supplied by the catalog.

Streaming

Yes

Tool calling

Yes

Structured outputs

Yes

Reasoning

Yes

OpenAI-compatible chat completions

Yes

JSON responses

Yes

Batch API

No

Fine-tuning

No

Dedicated deployment

No

Catalog precision: fp8. Maximum output: 393.22K tokens.

From model ID to request

Use DeepSeek V4 Flash with the OpenAI SDK

Install the SDK with npm install openai or pip install openai. Set GEODD_API_KEY in your server environment or secret manager. Never put it in browser code or a public environment variable.

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.GEODD_API_KEY,
  baseURL: "https://api.geodd.io/inference/v1",
});

const response = await client.chat.completions.create({
  model: "deepseek-ai/DeepSeek-V4-Flash",
  messages: [
    { role: "user", content: "Explain speculative decoding." }
  ],
});

console.log(response.choices[0].message.content);

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["GEODD_API_KEY"],
    base_url="https://api.geodd.io/inference/v1",
)

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Flash",
    messages=[
        {"role": "user", "content": "Explain speculative decoding."}
    ],
)

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

curl --fail-with-body "https://api.geodd.io/inference/v1/chat/completions" \
  --header "Authorization: Bearer $GEODD_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "deepseek-ai/DeepSeek-V4-Flash",
    "messages": [
      { "role": "user", "content": "Explain speculative decoding." }
    ]
  }'

All examples use deepseek-ai/DeepSeek-V4-Flash. To select another available region, replace baseURL, base_url, or the cURL URL with its base URL below. A real request consumes billable usage.

Usage-based billing / USD

DeepSeek V4 Flash pricing

Serverless billing unit

Per token; prices shown per 1 million tokens

Per-request fee

No per-request fee listed ($0)

Cache rates are catalog fields; a zero rate does not establish that caching is available. Cache support and reasoning-token accounting have not been verified. Confirm unlisted billing conditions before production use.

Model-specific availability

Available regions

Use the selected serving region for inference processing. Residency commitments and administrative or support access remain subject to Geodd's data handling policy and your agreement.

Payloads and operational metadata

Data handling

GDPR

Ready

SOC 2 Type II

Pending

DPA

Available

• Zero data retention by default. Prompts, outputs, and inference request bodies are processed transiently and are not stored unless separately agreed in writing.
• Geodd does not use customer prompts or model outputs to train models.
• Limited operational metadata, such as usage, billing, and security events, may be retained. Zero payload retention does not mean zero metadata.
• Requests use the selected serving region where applicable. API credentials should only be used server-side.

Integrate in your existing stack

Add DeepSeek V4 Flash with your coding agent

Sign in from your terminal, then give the prompt below to Claude Code, Codex, Cursor, or your preferred coding agent. Requires Node.js 22+ and npm; no global install needed.

npx @geodd/cli@latest auth login

Complete Google sign-in and any two-factor verification in the browser on the same computer. New to Geodd? Use auth signup instead of auth login.

Your agent uses the CLI to find the model's key-command ID and create a key with your approval, then configures your app. New keys use PostPaid billing with a default monthly capacity of 3 billion tokens, not a spending cap. Keep key secrets out of chat and shared logs.

Add DeepSeek V4 Flash to this application using Geodd.

Use model: deepseek-ai/DeepSeek-V4-Flash
Use Geodd's OpenAI-compatible API.

Requirements:
- Inspect the existing application first and use its existing integration patterns.
- Keep Geodd credentials server-side and use environment variables.
- Never expose the API credential to browser code, logs, or source control.
- Use GEODD_API_KEY; do not ask me to paste a credential into the conversation.
- Implement streaming where supported and appropriate.
- Preserve conversation history if the application includes chat.
- Add error handling for authentication, rate limits, and upstream failures.
- Install the required SDK using the project's package manager.
- Run a real test request when authorized credentials and billing are available.
- Ask for approval before incurring charges or changing a spending limit.
- If credentials are unavailable, explain the blocker; do not claim the test passed.
- Do not rewrite the application architecture when the existing stack supports this integration.

Geodd API base URL: https://api.geodd.io/inference/v1
Geodd documentation: https://geodd.io/docs/llm-api/getting-started
Model facts: https://geodd.io/models/deepseek-ai/DeepSeek-V4-Flash.json

CLI setup (Node.js 22+ and npm; no global install needed):
- Read https://geodd.io/docs/llm-api/cli-quick-start before using the CLI.
- Check the session with npx @geodd/cli@latest auth status. If needed, ask me to run npx @geodd/cli@latest auth login (or auth signup for a new account) and complete browser sign-in and two-factor verification on the same computer.
- Reuse an authorized GEODD_API_KEY if already configured. Otherwise, run npx @geodd/cli@latest models list --for-keys --json and identify the entry for deepseek-ai/DeepSeek-V4-Flash. Use that entry's id for key commands, not the public inference ID or display name. If the match is unclear, stop and ask.
- Before creating a key, get my approval for PostPaid billing and monthly capacity. The default is 3,000,000,000 tokens, not a spending cap; use --monthly-volume for an approved override.
- Choose a globally unique key name of 1-32 letters, numbers, or dashes. Run npx @geodd/cli@latest keys create --name KEY_NAME --model MODEL_ID --json with the chosen name and discovered key-command ID. Add --yes only after explicit approval for a non-interactive operation.
- Capture secret-bearing output privately, never in shared logs or the conversation. Store data.data.apiKey securely as server-side GEODD_API_KEY and retain data.data.keyId for updates. The CLI does not save the key secret. If private capture is unavailable, ask me to provision the secret outside the agent session. Do not blindly retry creation after a timeout.
- There is no keys list command. If updating a saved keyId, keys update replaces the entire model set; include every model to retain and get approval first.
- Keep deepseek-ai/DeepSeek-V4-Flash as the model ID in inference requests. The coding agent, not the CLI, must configure the application and run the approved test request.

Traceable facts

Model data and sources

This page and its JSON representation use the same backend snapshot. Unknown values remain null in JSON; they are not interpreted as supported or unsupported. The availability date uses the backend's model creation timestamp.

Model name

DeepSeek V4 Flash

Model ID

deepseek-ai/DeepSeek-V4-Flash

Model developer

deepseek-ai

API provider

Geodd

Last verified (catalog)

October 3, 2026

Page data updated

October 3, 2026

Model released

Not supplied by the Geodd catalog

Available on Geodd since

May 19, 2026

Canonical model page

https://geodd.io/models/deepseek-ai/DeepSeek-V4-Flash

Direct answers

Frequently asked questions