Skip to main content

Migrating from Anthropic to Turing Platform

To migrate code using the Anthropic Python/TypeScript SDK to the Turing Platform, you need to change three things: base_url, the authentication header type (x-api-key → Authorization: Bearer), and the model name.

Core diff​

Python​

 from anthropic import Anthropic

 client = Anthropic(
-    api_key=os.environ["ANTHROPIC_API_KEY"],
+    auth_token=os.environ["TURING_API_KEY"],
+    base_url="https://live-turing.cn.llm.tcljd.com/api/v1",
 )

 message = client.messages.create(
-    model="claude-sonnet-5",
+    model="turing/claude-sonnet-5",
     max_tokens=1024,
     messages=[{"role": "user", "content": "Hello"}],
 )
api_key → auth_token

The Anthropic SDK authenticates using the x-api-key header by default. The Turing Platform uses the standard Authorization: Bearer header — in the SDK, this corresponds to the auth_token parameter (not api_key).

TypeScript / Node.js​

 import Anthropic from "@anthropic-ai/sdk";

 const client = new Anthropic({
-  apiKey: process.env.ANTHROPIC_API_KEY,
+  authToken: process.env.TURING_API_KEY,
+  baseURL: "https://live-turing.cn.llm.tcljd.com/api/v1",
 });

cURL​

-curl https://api.anthropic.com/v1/messages \
+curl https://live-turing.cn.llm.tcljd.com/api/v1/messages \
-  -H "x-api-key: $ANTHROPIC_API_KEY" \
-  -H "anthropic-version: 2023-06-01" \
+  -H "Authorization: Bearer $TURING_API_KEY" \
   -H "Content-Type: application/json" \
   -d '{
-    "model": "claude-sonnet-5",
+    "model": "turing/claude-sonnet-5",
     "max_tokens": 1024,
     "messages": [{"role": "user", "content": "Hello"}]
   }'
anthropic-version header

The Turing Platform automatically selects the appropriate API version internally. You may keep this header (it is harmless) or remove it.

Model name mapping​

Claude model names on the Turing Platform are prefixed with turing/claude-*. When migrating, prefer recent model versions such as turing/claude-sonnet-5 or turing/claude-opus-4.8.

For the full list of available models, context windows, pricing, and retirement status, refer to Model List → Turing Claude.

Feature comparison​

FeatureAnthropic NativeTuring Platform
Messages API✅✅ Fully compatible; see Messages API
Streaming SSE✅✅
Function calling (tools + tool_result)✅✅
Extended Thinking✅ {"type": "enabled"/"adaptive"}✅
Prompt Caching (cache_control)✅✅, with 5m / 1h TTL
Vision (image input)✅✅
Web Search Tool (web_search_20250305)✅✅
Computer Use✅⚠ Routed via upstream Vertex AI; supported in some scenarios
Files API✅⚠ Planned
Batch API✅⚠ Not publicly available
Count Tokens✅✅ /api/count-message-tokens

Turing Platform addition: turing_options​

Add a turing_options field at the top level of the request to access platform capabilities:

message = client.messages.create(
    model="turing/claude-sonnet-5",
    max_tokens=1024,
    messages=[...],
    extra_body={
        "turing_options": {
            "timeout": 30,
            "max_retries": 2,
            "fallbacks": ["turing/gpt-5.4-mini"]
        }
    }
)

See Timeout, Retry, and Fallback.

Common gotchas​

  • Using api_key by mistake: The SDK will automatically use the x-api-key header, causing the Turing Platform to return 401; you must use auth_token.
  • Model name date suffixes: Date-suffixed names like claude-sonnet-4-5-20241022 are not used on the Turing Platform — dates are dropped: turing/claude-sonnet-4.5.
  • anthropic-version header: No need to set this explicitly.
  • Response fields id, type, role, etc.: Identical to Anthropic's responses; no changes to parsing logic are required.
  • Thinking configuration for newer Claude models: claude-opus-4.7 / claude-opus-4.8 / claude-sonnet-5 do not support type: "enabled" + budget_tokens; use type: "adaptive" + output_config.effort instead. Sonnet 5 has reasoning enabled by default but can be disabled; Opus 4.7/4.8 has it disabled by default and must be explicitly enabled. See Reasoning for details.

Optional: Switch to the Chat Completions protocol​

The Anthropic Messages protocol is Claude's native protocol and is not portable across providers. If you want to use the same code to switch freely between Claude, GPT, Gemini, and Qwen, switch to the OpenAI Chat Completions protocol:

-from anthropic import Anthropic
-client = Anthropic(
-    auth_token="...",
-    base_url="https://live-turing.cn.llm.tcljd.com/api/v1",
-)
+from openai import OpenAI
+client = OpenAI(
+    api_key="...",
+    base_url="https://live-turing.cn.llm.tcljd.com/api/v1",
+)

-message = client.messages.create(
-    model="turing/claude-sonnet-5",
-    max_tokens=1024,
-    messages=[{"role": "user", "content": "Hello"}],
-)
+response = client.chat.completions.create(
+    model="turing/claude-sonnet-5",  # Change only this line to switch models
+    messages=[{"role": "user", "content": "Hello"}],
+)
When to stay on the Messages protocol

If you need Claude-native features — such as fine-grained cache_control caching, server_tool_use / web_search_tool_result block structures, or direct return of thinking blocks — staying on the Messages protocol provides a better experience.

Platform capabilities available immediately after migration​

  • Fallback to non-Claude models: If claude-sonnet-5 fails, automatically switches to gpt-5.4-mini.
  • Unified billing view for cross-provider prompt caching: Billing & Usage.
  • Request tracing: X-Turing-Trace-Id response header.
  • Unified rate limit view: Rate Limits.

See also​