Migrating from OpenAI

Already using the OpenAI SDK? Switching takes two lines, and then you can call 90+ models — not just GPT.

No sk-gpushare-* key yet? Create one at dianqi.zsopc.com/dashboard/keys. Sign-up includes $0.30 of trial credit, enough for every example here.

What changes (Python)#

from openai import OpenAI

# straight to OpenAI, before
client = OpenAI(
-    api_key="sk-...",
+    api_key="sk-gpushare-xxx",
+    base_url="https://dianqi.zsopc.com/v1",
)

That's both changes. Everything after client.chat.completions.create(...) stays exactly as it is.

What changes (TypeScript)#

import OpenAI from "openai";

const client = new OpenAI({
-  apiKey: process.env.OPENAI_API_KEY,
+  apiKey: process.env.PLATFORM_API_KEY,
+  baseURL: "https://dianqi.zsopc.com/v1",
});

What changes (curl)#

- curl https://api.openai.com/v1/chat/completions \
+ curl https://dianqi.zsopc.com/v1/chat/completions \
-   -H "Authorization: Bearer sk-..." \
+   -H "Authorization: Bearer sk-gpushare-xxx" \
    -H "Content-Type: application/json" \
    -d '{"model":"gpt-5.4","messages":[{"role":"user","content":"Hello"}]}'

What you gain#

Once the base URL is switched, your code can call more than GPT:

# same client, same code, different model
client.chat.completions.create(model="gpt-5.4", ...)               # still works
client.chat.completions.create(model="claude-sonnet-4-6", ...)     # new!
client.chat.completions.create(model="gemini-2.5-pro", ...)        # new!
client.chat.completions.create(model="glm-5.1", ...)               # new!
client.chat.completions.create(model="grok-4-fast-reasoning", ...) # new!

Full model IDs: Models.

Transparent pricing#

Pricing is public (see the model plaza), and the GPT line costs the same as going direct (as of June 2026):

Every call is itemised at dianqi.zsopc.com/dashboard/usage, matching what you'd see from the vendor; keys are managed at dianqi.zsopc.com/dashboard/keys.

Billing is a prepaid account wallet: all keys share one balance, and when it runs out every key stops working at once (HTTP 402). Top up at dflop.top/dashboard/billing (Stripe, $1 minimum, same account and shared balance as dianqi.zsopc.com).

Read this after switching#

1. Tool behaviour differs slightly#

• function tools are fully compatible and behave exactly as they do against OpenAI
• The web_search tool works with gpt-5.4 / gpt-5.5 and some Claude and Gemini models (full list in the compatibility matrix and the capability columns of Models), and requires stream=true
• The image_generation tool is the same — stream=true required
• Careful when mixing: as soon as web_search or image_generation appears in tools, the request takes a different translation path, and in multi-turn history the tool_calls field on assistant messages is not replayed (the gateway keeps only the assistant text; tool_call_id on tool-role messages still passes through). Pure function tools are unaffected — if you rely heavily on multi-turn function calls, avoid mixing in the built-in tools
• Details: Tool calling

2. Streaming is identical to OpenAI#

stream=true and the same SSE protocol. Your client parsing code doesn't change. See Streaming.

3. Error format is identical#

Errors from /v1/chat/completions keep the OpenAI shape:

{"error": {"message": "...", "type": "...", "code": "..."}}

4. Which OpenAI endpoints are supported#

If your app depends on TTS transcription, that part still needs the original vendor.

The platform also offers endpoints OpenAI doesn't: async video generation (billed per second — see Image / video / music APIs) and knowledge-base search over REST and MCP (see Knowledge base API & MCP).

Full before/after#

Straight to OpenAI#

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Write a haiku"}],
)
print(resp.choices[0].message.content)

Through this platform#

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["PLATFORM_API_KEY"],
    base_url="https://dianqi.zsopc.com/v1",
)

# still GPT
resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Write a haiku"}],
)

# and now Claude too
resp_claude = client.chat.completions.create(
    model="claude-sonnet-4-6",
    messages=[{"role": "user", "content": "Write a haiku"}],
)

FAQ#

Should I delete my OPENAI_API_KEY env var?#

No need — keep both around:

export OPENAI_API_KEY=sk-...           # straight to OpenAI
export PLATFORM_API_KEY=sk-gpushare-...

And pick explicitly in code:

# fall back to the vendor when it matters
client = OpenAI(
    api_key=os.environ.get("PLATFORM_API_KEY") or os.environ["OPENAI_API_KEY"],
    base_url=os.environ.get("OPENAI_BASE_URL", "https://api.openai.com/v1"),
)

Are the rate limits the same as OpenAI's?#

No. There's no enforced QPS ceiling here (occasional 429s are upstream rate limits passed through), and billing is a prepaid account wallet: all keys share one balance, and when it's exhausted every key returns HTTP 402 (type insufficient_quota, code quota_exceeded). A new key brings no new credit — topping up does (dflop.top/dashboard/billing). See Authentication and Error codes.

Do I need to change my retry logic?#

No. The OpenAI SDK's built-in exponential backoff works here too, and status-code semantics are consistent across services (Error codes).

Do function-tool tool_call_ids change between calls?#

On the pure function tool path the gateway passes upstream IDs through without rewriting, so your multi-turn tool-result code needs no changes. But if the same request mixes in the web_search or image_generation built-ins, assistant tool_calls in history aren't replayed — see Tool behaviour differs slightly above.

Next steps#

• Which model gives the best value? See Models § picking by job
• Anthropic SDK or OpenAI SDK for Claude? Personal preference — both work, use the one you know
• Client integrations: Cursor / Cline / Continue