DeepSeek V4 Flash

DeepSeek V4 Flash is a general-purpose open-weight model served on Sarvam infrastructure. It supports text input and output with a 1,048,576-token context window.

At a glance

Key capabilities

Quickstart

from sarvamai import SarvamAI
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
response = client.chat.completions_v2(
    model="deepseekv4-flash",
    messages=[
        {
            "role": "user",
            "content": "Explain the difference between a B-tree and a B+ tree index.",
        }
    ],
    temperature=1,
    top_p=0.95,
    max_tokens=2000,
)
print(response.choices[0].message.content)

API reference

Call POST /v2/chat/completions or POST /v2/responses with model: "deepseekv4-flash". Request fields are defined in those API references — not duplicated on this model page.

Reasoning

DeepSeek V4 Flash has three reasoning modes — low, high, and max — and accepts all three directly on both APIs:

This is different from GLM-5.3, where the Responses API rejects the literal string "max" — DeepSeek V4 Flash has no such restriction.

Read the reasoning trace from choices[].message.reasoning_content on chat, or from the reasoning item’s summary[0].text in output on Responses when one is returned. Reasoning tokens are billed as output tokens either way, and count against max_tokens on chat (defaults to 2048 when omitted) or max_output_tokens on Responses (no default — omitting it means unbounded generation).

To turn reasoning off on chat, set extra_body.chat_template_kwargs.enable_thinking to false.

Known limitations

• Reasoning and the final answer share the max_tokens / max_output_tokens budget. Raise the cap or disable thinking via extra_body.chat_template_kwargs.enable_thinking: false when needed.
• Structured output with reasoning enabled can return content: null — disable thinking or raise the token cap.

Limits and errors