DeepSeek API access, OpenAI-compatible — and the reasoning_content trap

DeepSeek is one of the best value-for-money families, and it's often painful to sign up for from abroad — through DaoXE it's behind the same key as GPT and Claude. Call /v1/chat/completions with a deepseek-* ID. See the cheapest APIs guide for how it fits a cheap/frontier split.

Call DeepSeek through DaoXE#

Use /v1/chat/completions with an exact deepseek-* ID. There are non-thinking and thinking variants; the thinking variant returns a separate reasoning_content field.

curl --fail-with-body --show-error --silent \
  https://daoxe.com/v1/chat/completions \
  -H "Authorization: Bearer ${DAOXE_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "YOUR_EXACT_MODEL_ID",
    "max_tokens": 64,
    "messages": [{"role": "user", "content": "Say hello in one sentence."}]
  }'

# DeepSeek "thinking" variants return a separate reasoning_content field.
# In multi-turn tool use you MUST echo the previous assistant message back
# (reasoning_content included) or the API answers 400.
from openai import OpenAI

client = OpenAI(base_url="https://daoxe.com/v1", api_key="YOUR_DAOXE_KEY")
r = client.chat.completions.create(
    model="YOUR_EXACT_MODEL_ID",   # a deepseek-* id from GET /v1/models
    messages=[{"role": "user", "content": "2+2? Think briefly, then answer."}],
)
msg = r.choices[0].message
print(getattr(msg, "reasoning_content", None))  # chain-of-thought, if exposed
print(msg.content)                               # final answer

How DeepSeek exposes reasoning#

DeepSeek R-series IDs put the chain of thought in a dedicated reasoning_content field on the message, separate from content. The trap is multi-turn tool use: the API requires the previous assistant message to be echoed back including its reasoning_content, or the request 400s. When you build history, build it from the raw message objects, not from a trimmed summary of the text.

DeepSeek-specific notes#

• The reasoning_content round-trip. With a thinking variant in multi-turn tool use, you MUST echo the previous assistant message (including reasoning_content) back, or the API answers 400.
• Non-thinking needs no round-trip. If you don't need visible reasoning, use the non-thinking variant and avoid the trap entirely.
• Frameworks can drop the field. Some clients silently discard reasoning_content — check yours before enabling thinking.
• Exact IDs only. Pick the right variant by its ID from GET /v1/models.

What it costs#

Verify you actually get the model#

Prove the endpoint works before blaming the client — if this fails, no setting will fix it:

export DAOXE_API_KEY="your_api_key"

# List the exact model IDs your account can call
curl --fail-with-body --show-error --silent \
  https://daoxe.com/v1/models \
  -H "Authorization: Bearer ${DAOXE_API_KEY}"

Confirm connectivity, then diff a hard reasoning prompt against the official DeepSeek API at temperature 0:

Frequently asked questions#