DeepSeek family

DeepSeek’s V4 series models, including V4.1 Flash, V4 Pro and V4 Flash. They are called through the OpenAI-compatible API and also support the Anthropic-native API, so they work directly in tools such as Claude Code.

Common tasks

Models

The V4 series has a 1M-token context and a maximum output of 384K tokens. V4.1 Flash accepts image input; the other models are text-only. For the full model list and live prices, see the model page .

Aliases and versions

• Use aliases without the deepseek/ prefix, e.g. "model": "deepseek-v4-pro"; deepseek/deepseek-v4-pro returns 404.
• deepseek-chat and deepseek-reasoner from DeepSeek’s official API are not available on OfoxAI; use the model IDs in the table above instead.
• The earlier versions deepseek/deepseek-v4-pro-0423 and deepseek/deepseek-v4-flash-0423 can still be called, but enable thinking differently from the newer versions — see Thinking mode.

Examples

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])
 
resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-flash-0731",
    messages=[{"role": "user", "content": "Explain idempotency in one sentence."}],
    extra_body={"thinking": {"type": "disabled"}},
)
print(resp.choices[0].message.content)

curl https://api.ofox.ai/v1/chat/completions \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash-0731",
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}],
    "thinking": {"type": "disabled"}
  }'

Thinking mode

In our tests on the same math problem, V4 Pro thought for about 12,000 tokens by default and took about 2 minutes; with thinking off it took about 20 seconds. Thinking tokens are billed at the output rate; the count is in usage.completion_tokens_details.reasoning_tokens.

Reading the thinking content

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])
 
resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-flash-0731",
    messages=[{"role": "user", "content": "What is 17 × 23?"}],
)
msg = resp.choices[0].message
details = getattr(msg, "reasoning_details", None) or []
print("[Thinking]", "".join(d.get("text", "") for d in details))
print("[Answer]", msg.content)

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])
 
stream = client.chat.completions.create(
    model="deepseek/deepseek-v4-flash-0731",
    messages=[{"role": "user", "content": "What is 17 × 23?"}],
    stream=True,
)
for chunk in stream:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta
    thinking = getattr(delta, "reasoning_content", None)
    if thinking:
        print(thinking, end="", flush=True)
    if delta.content:
        print(delta.content, end="", flush=True)

• In non-streaming responses, thinking content is in message.reasoning_details; concatenate the text of each item in order.
• In streaming responses, thinking content is in delta.reasoning_content, the same as DeepSeek’s official API.
• In multi-turn conversations, you don’t need to send thinking content back in the message history.

Use in Claude Code

The DeepSeek V4 series supports the Anthropic-native API at https://api.ofox.ai/anthropic, so it works in Claude Code and other tools built on the Anthropic protocol:

curl https://api.ofox.ai/anthropic/v1/messages \
  -H "x-api-key: $OFOX_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash-0731",
    "max_tokens": 4000,
    "thinking": {"type": "disabled"},
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}]
  }'

• On the Anthropic API, thinking content is returned as thinking blocks, and "thinking": {"type": "disabled"} also turns thinking off.
• Set a large enough max_tokens when thinking is on. If the budget runs out, the response may contain only a thinking block and no answer.

For full Claude Code setup steps, see Claude Code setup.

JSON output and tool calling

response_format supports {"type": "json_object"} and {"type": "json_schema", ...}; tool calling uses the standard tools parameter:

import json
import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])
 
resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro-0813",
    response_format={"type": "json_object"},
    messages=[{"role": "user", "content": "Return price and currency as JSON: A bat and a ball cost $1.10 in total. The bat costs $1 more than the ball. How much does the ball cost?"}],
    extra_body={"thinking": {"type": "disabled"}},
)
print(json.loads(resp.choices[0].message.content))

When using json_object, state “return JSON” and the required fields in the prompt.

Prompt caching

DeepSeek automatically caches repeated request prefixes, with no changes to your request. The cached amount is in usage.prompt_tokens_details.cached_tokens, and the cached portion is billed at the cache read rate. Caching is best-effort; requests with the same prefix may still miss.

Providers

For how to select a provider, see Provider routing.

Common errors

For other general errors, see Error Handling.

Official docs

• DeepSeek API docs 
• DeepSeek · Thinking mode 
• DeepSeek · Anthropic API compatibility 

The official docs describe behavior when calling DeepSeek directly. When calling through OfoxAI, model IDs and endpoints on this page take precedence.

FAQ