DeepSeek V4 Flash

BudgetOpen weightsapac

DeepSeekChatdeepseek-v4-flash

Cheapest strong general-purpose model — 1M context, dual thinking/non-thinking modes.

Dual-mode (thinking/non-thinking) in one id; replaces deepseek-chat and deepseek-reasoner (both retire 2026-07-24).

Capabilities

Tool callingJSON mode

Recommended for

cost-sensitive workloads
high-volume generation

Using this model through Conduix

Call it by its id, or let a routing alias pick it as part of a governed, cross-provider chain with automatic failover.

Direct model iddeepseek-v4-flash

Reached by these routing aliases:

frontier-budgetfallback #3 · GPT-5.4 mini → Claude Haiku 4.5 → Gemini 3.1 Flash-Lite → DeepSeek V4 Flash
reasoning-budgetfallback #1 · DeepSeek V4 Pro → DeepSeek V4 Flash → GPT-5.4 mini

Routing relationships are operational — chains adapt to provider availability and your org's policy — not a contractual guarantee that a specific model serves a given request.

Get started

Conduix is OpenAI-API-compatible — point any OpenAI SDK at https://api.conduix.ai/v1 and pass deepseek-v4-flash as the model.

from conduix import Conduix

client = Conduix(api_key="cx_live_YOUR_KEY")  # base URL is built in

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello, Conduix!"}],
)
print(response.choices[0].message.content)

Tool calling: Pass a tools array and the model calls your functions.
JSON mode: Set response_format for guaranteed-valid JSON output.

The response's x-conduix-model-served header tells you exactly which model answered. See the quickstart for keys, headers, and usage.