Gemini 3.8 Flash

​

Key capabilities

• 1M token context - Up to 1,048,576 input tokens and 65,536 output tokens
• Multimodal input - Text, image, video, audio, and PDF input; text output
• Long-horizon coding - Designed for complex, multi-file engineering work and iterative verification
• Autonomous agents - Improved multi-step planning, deterministic tool execution, and recovery from failed loops
• Tools - Function calling, code execution, File Search, Google Search, Google Maps, URL context, structured output, and Computer Use (Preview)
• Thinking levels - low, medium, and high; the default is medium
• Caching and processing options - Supports context caching, Batch API, Flex inference, and Priority inference upstream

​

Model specifications

​

Quick example

curl "https://www.anyfast.ai/v1beta/models/gemini-3.8-flash:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [{ "text": "Review this retry design, identify race conditions, and propose a safe implementation plan." }]
      }
    ],
    "generationConfig": {
      "thinkingConfig": {
        "thinkingLevel": "medium"
      },
      "maxOutputTokens": 2048
    }
  }'

import requests

response = requests.post(
    "https://www.anyfast.ai/v1beta/models/gemini-3.8-flash:generateContent",
    params={"key": "YOUR_API_KEY"},
    json={
        "contents": [{
            "role": "user",
            "parts": [{
                "text": "Review this retry design, identify race conditions, and propose a safe implementation plan."
            }],
        }],
        "generationConfig": {
            "thinkingConfig": {"thinkingLevel": "medium"},
            "maxOutputTokens": 2048,
        },
    },
)
response.raise_for_status()
print(response.json()["candidates"][0]["content"]["parts"][0]["text"])

​

Thinking levels

​

Removed generation parameters

• generationConfig.temperature
• generationConfig.topP
• generationConfig.topK
• generationConfig.candidateCount
• generationConfig.thinkingConfig.thinkingBudget

​

Conversation and tool requirements

• Do not prefill the model response. The final conversation turn must be a non-empty user turn.
• Preserve every thoughtSignature returned by the model when sending conversation history back in a tool workflow.
• A functionResponse must include both the function name and the matching call id returned by the model.
• Keep multimodal assets inside their corresponding content or function-response payload.

​

Parameters

​

Official references

• Gemini 3.8 Flash model page
• What’s new in Gemini 3.8 Flash
• Gemini thinking