Skip to main content
deepseek-v3-anthropic is DeepSeek’s V3.2 model accessed via the Anthropic Messages protocol at POST /v1/messages. The model weights are identical to deepseek-v3-2 — the differentiator is the wire format. If your code already speaks Anthropic Messages (DOSIA Agent mode, anthropic SDK, any tool_use / cache_control / thinking-block consumer), this is the alias to point it at; you don’t have to rewrite the call site. Pricing: 0.14/1Minput,0.14 / 1M input, 0.28 / 1M output, $0.003 / 1M cache read — see the rate card. See the Anthropic Messages protocol for what passes through.

​
Protocols

ProtocolPath
Anthropic MessagesPOST https://llm.bytespike.ai/v1/messages
OpenAI Chat Completions is not wired on this alias — use deepseek-v3-2 if you need the OpenAI-shape interface.

​
Quickstart

curl https://llm.bytespike.ai/v1/messages \
  -H "x-api-key: $BYTESPIKE_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "deepseek-v3-anthropic",
    "max_tokens": 1024,
    "messages": [
      { "role": "user", "content": "Summarize this changelog into three bullets." }
    ]
  }'

​
Capabilities

CapabilitySupported
Chat completions✅
Streaming (SSE)✅
Tool use (function calling)✅ parallel
Prompt caching (cache_control)✅
Extended thinking— (use deepseek-r1 for reasoning chains)
Vision (image input)—
JSON / structured output✅
Context window64K tokens

​
When to use

  • You already speak Anthropic Messages — DOSIA Agent mode, anthropic SDK, a tool-use-heavy loop. Drop in the alias, save the rewrite.
  • Cost-sensitive Anthropic-shape workloads — DeepSeek V3.2 is a small fraction of the input price of claude-opus-4-8 and one-twentieth of claude-sonnet-4-6, while keeping the same wire format.
  • Tool-use mixed batches — when some of your agents target Claude and some target DeepSeek but everyone speaks the same protocol.
When not to use:
  • You need vision input — use a Claude family model or gemini-3-1-pro.
  • You’re already on the OpenAI Chat Completions interface and don’t want to migrate — stay on deepseek-v3-2.
  • You need explicit reasoning chains — try deepseek-r1.

​
Next