Gemini (Google-Compatible)

BeLive Studio provides Google-native Gemini endpoints, usable with the official Google GenAI SDK or by pointing Gemini CLI at BeLive Studio.

Endpoints

POST https://api.tokenhub.belivestudios.com/v1beta/models/{model}:generateContent
POST https://api.tokenhub.belivestudios.com/v1beta/models/{model}:streamGenerateContent

Request Example

curl "https://api.tokenhub.belivestudios.com/v1beta/models/gemini-3.7-flash:generateContent" \
  -H "x-goog-api-key: sk-your-belive-studios-key" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "Hello"}]
    }]
  }'

Common request parameters

Output tokens and thinking

The Gemini-native interface uses generationConfig.maxOutputTokens to set the output limit; it does not use the OpenAI-compatible max_tokens field. For models that support thinking, reasoning also consumes the output budget. If the limit is too small, the API may return HTTP 200 with no displayable text and candidates[0].finishReason set to "MAX_TOKENS".

{
  "contents": [{"parts": [{"text": "Introduce BeLive Studio in one sentence"}]}],
  "generationConfig": {"maxOutputTokens": 1024}
}

Do not judge success from the HTTP status alone. Check candidates[].content.parts and candidates[].finishReason. If there is no text and the finish reason is "MAX_TOKENS", increase maxOutputTokens and retry, then inspect usageMetadata.thoughtsTokenCount to determine whether reasoning consumed most of the budget. There is no single minimum that applies to every Gemini model.

The following low-budget response was tested on the platform:

{
  "candidates": [{"finishReason": "MAX_TOKENS"}],
  "usageMetadata": {
    "promptTokenCount": 7,
    "totalTokenCount": 13,
    "thoughtsTokenCount": 6
  }
}

Models that support thinking configuration use generationConfig.thinkingConfig. The control for disabling thinking differs by model. Follow the compatibility table below rather than inferring support from the model ID.

Disabling thinking

The following results were verified through the BeLive Studio Google-native endpoint on September 10, 2026. Every test request used maxOutputTokens: 2000:

For the first three models, use the following request and replace {model} in the URL with the model ID:

curl "https://api.tokenhub.belivestudios.com/v1beta/models/{model}:generateContent" \
  -H "x-goog-api-key: sk-your-belive-studios-key" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "Analyze this problem and provide the answer"}]
    }],
    "generationConfig": {
      "maxOutputTokens": 2000,
      "thinkingConfig": {
        "thinkingBudget": 0
      }
    }
  }'

Check usageMetadata.thoughtsTokenCount in the response. If it is 0 or omitted, that request did not consume thinking tokens.

To request returnable reasoning content, set includeThoughts: true. When the model produces returnable reasoning, the response contains parts marked with thought: true. Applications should keep these separate from the final answer:

{
  "contents": [{"parts": [{"text": "Analyze 2+2 and give the answer"}]}],
  "generationConfig": {
    "maxOutputTokens": 512,
    "thinkingConfig": {
      "thinkingLevel": "LOW",
      "includeThoughts": true
    }
  }
}

includeThoughts only controls whether returnable reasoning content is included. It does not disable thinking or guarantee lower thinking-token usage. Available thinking fields and values depend on the model version; sending thinking configuration to an unsupported model returns an error.

Response fields

Streaming

streamGenerateContent returns text/event-stream, with each event prefixed by data::

curl -N "https://api.tokenhub.belivestudios.com/v1beta/models/gemini-3.7-flash:streamGenerateContent" \
  -H "x-goog-api-key: sk-your-belive-studios-key" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "Write a two-line poem"}]}]
  }'

Parse the JSON following each data: prefix. Intermediate events may omit finishReason; the final event provides the finish reason and usage information.

Python SDK

from google import genai

client = genai.Client(
    api_key="sk-your-belive-studios-key",
    http_options={"base_url": "https://api.tokenhub.belivestudios.com"},
)

resp = client.models.generate_content(
    model="gemini-3.7-flash",
    contents="Hello",
)
print(resp.text)

The SDK appends /v1beta/models/... automatically, so set base_url to the host only and do not add /v1beta.

Client Integration

Gemini CLI connects via the GOOGLE_GEMINI_BASE_URL and GEMINI_API_KEY environment variables — see the Gemini CLI guide.