# Gemini thinking settings: thinkingLevel, thinkingBudget and reasoning_effort

Source: https://ofox.ai/docs/develop/models/gemini/thinking

This page covers Gemini's thinking defaults and how to adjust them. For model IDs and endpoints, see the [Gemini family overview](https://ofox.ai/docs/develop/models/gemini).

**Key point**:Except for Flash-Lite, Gemini models think by default, and complex questions can take more than a minute. Reduce thinking when speed matters.

Gemini models think before answering. Thinking tokens count as output and are billed at the output rate; the count is in `usageMetadata.thoughtsTokenCount` (`usage.completion_tokens_details.reasoning_tokens` on the OpenAI-compatible API).

| Model | Default | How to adjust |
|---|---|---|
| Gemini 3.x (except Flash-Lite) | Thinking on | `thinkingLevel`, e.g. `"low"`; `thinkingBudget` (a cap on thinking tokens) is also accepted |
| Flash-Lite series (3.5, 3.1) | No thinking | Enabled by passing `thinkingLevel` or `thinkingBudget` |

In our tests on the same math problem, `thinkingLevel: "low"` cut Gemini 3.5 Flash's thinking tokens from about 6000 to about 4600, with a corresponding drop in latency. `thinkingBudget: 0` turns off thinking on 3.5 Flash and 3 Flash Preview; 3.1 Pro Preview and 3.8 Flash still think when it is passed and cannot turn thinking off completely.

**Python (google-genai)**

```python filename="gemini_thinking.py"
import os
from google import genai
from google.genai import types

client = genai.Client(
    api_key=os.environ["OFOX_API_KEY"],
    http_options={"api_version": "v1beta", "base_url": "https://api.ofox.ai/gemini"},
)

resp = client.models.generate_content(
    model="google/gemini-3.8-flash",
    contents="Find all positive integers n less than 50 for which n² + n + 41 is not prime.",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_level="low"),
    ),
)
print(resp.text)
print("Thinking tokens:", resp.usage_metadata.thoughts_token_count or 0)
```

**cURL**

```bash filename="Terminal"
curl "https://api.ofox.ai/gemini/v1beta/models/google/gemini-3.8-flash:generateContent" \
  -H "x-goog-api-key: $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "Find all positive integers n less than 50 for which n² + n + 41 is not prime."}]}],
    "generationConfig": {
      "thinkingConfig": {"thinkingLevel": "low"}
    }
  }'
```

**Python (OpenAI SDK)**

```python filename="gemini_thinking_openai.py"
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])

resp = client.chat.completions.create(
    model="google/gemini-3.8-flash",
    reasoning_effort="low",
    messages=[{"role": "user", "content": "Find all positive integers n less than 50 for which n² + n + 41 is not prime."}],
)
print(resp.choices[0].message.content)
details = resp.usage.completion_tokens_details
print("Thinking tokens:", details.reasoning_tokens if details else 0)
```

- With `thinkingLevel: "low"`, the model may skip thinking on simple questions, in which case `thoughtsTokenCount` does not appear.
- To see the thinking process, add `"includeThoughts": true` to `thinkingConfig`. Items in `parts` with `"thought": true` are thought summaries; the rest is the answer.
- Thinking tokens count toward `maxOutputTokens`. If the limit is too low, the answer is cut off midway — see [Common errors](https://ofox.ai/docs/develop/models/gemini#errors).
- `thinkingConfig` goes inside `generationConfig`.