Gemini 3.8 Flash is Google’s generally available Flash model for long-horizon software engineering, autonomous agents and complex enterprise workflows. It uses the stable API ID gemini-3.8-flash, accepts text, images, video, audio and PDFs, and returns text with up to 65,536 output tokens.

The immediate attraction is price: $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Google has already scheduled the standard price to double on January 1, 2027, so production budgets should store both rates.

Gemini 3.8 Flash specifications

Specification	Gemini 3.8 Flash
Status	Generally available
Release date	September 2, 2026
API model ID	gemini-3.8-flash
Input limit	1,048,576 tokens
Output limit	65,536 tokens
Input	Text, image, video, audio and PDF
Output	Text
Thinking levels	Low, medium and high
Default thinking	Medium
Function calling	Supported
Structured output	Supported
Computer use	Supported in preview
Batch, Flex, Priority	Supported

Computer use remains a preview capability even though the model itself is GA. The model does not support the Live API, image generation or audio generation.

Pricing and effective dates

Consumption mode	Through Dec 31, 2026	From Jan 1, 2027
Standard input	$0.75/M	$1.50/M
Standard output, including thinking	$3.75/M	$7.50/M
Cached input	$0.075/M	$0.15/M
Batch input	$0.375/M	$0.75/M
Batch output	$1.875/M	$3.75/M
Flex input	$0.375/M	$0.75/M
Flex output	$1.875/M	$3.75/M
Priority input	$1.35/M	$2.70/M
Priority output	$6.75/M	$13.50/M

Output pricing includes thinking tokens. Cached-context storage is billed separately: Google lists $0.50 per million tokens per hour through December 31 and $1.00 from January 1.

The free tier is useful for development, but Google states that free-tier content may be used to improve its products. Paid-tier content is not used for that purpose. Verify account and regional terms before sending sensitive data.

Minimal API example

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3.8-flash",
    contents="Review this retry workflow and identify race conditions.",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_level="medium")
    ),
)

print(response.text)

Google recommends the current google-genai SDK. For ordinary calls, omit unsupported legacy sampling controls rather than copying an older Gemini configuration.

Thinking levels

Gemini 3.8 Flash supports three string-valued levels:

• low for latency-sensitive tasks and routine processing;
• medium, the default, for most coding and agent work;
• high for harder reasoning and multi-step orchestration.

minimal is not supported and returns an error. The old numeric thinking_budget field must be replaced with thinking_level. Because thinking tokens are billed as output, raising the level can change both latency and cost.

Tools and agent workflows

The model supports function calling, structured outputs, code execution, file search, URL context, Google Search grounding and Google Maps grounding. Google’s Managed Agents documentation says the Antigravity agent and Antigravity SDK now use Gemini 3.8 Flash by default.

That does not mean every third-party agent product automatically migrated. It is a Google Managed Agents default, not a universal Gemini ecosystem alias.

For production agents, validate tool arguments, cap loops and require approval for consequential actions. Our AI application architecture hub covers reliable orchestration, and AI Testing & Evaluation covers regression gates.

Migration from Gemini 3.7 or older

Changing only the model ID is not always sufficient. Google’s current migration checklist requires developers to:

1. change the model string to gemini-3.8-flash;
2. remove temperature, top_p and top_k;
3. replace numeric thinking_budget with thinking_level;
4. remove candidate_count;
5. remove prefilled model turns;
6. ensure the last user turn contains non-empty text;
7. include both function name and matching call_id in function responses;
8. place multimodal tool assets inside the function-response payload;
9. separate inline instructions with \n\n where Google requires that format.

Test a normal response, multi-turn conversation, function call, multimodal function response and each thinking level before moving traffic. Keep Gemini 3.7 Flash available as a rollback target during migration.

Gemini 3.7 versus 3.8

Both models currently share the same introductory and scheduled standard token prices. Gemini 3.8 is the newer production owner for long-horizon coding and autonomous agents, adds a documented 64K output ceiling and is Google’s default for Antigravity Managed Agents. Gemini 3.7 remains supported, so teams can migrate through controlled evaluation instead of an emergency cutover.

Google reports stronger coding, agent and enterprise-workflow performance for 3.8. Those are Google-attributed claims; we have not independently reproduced a universal improvement across every codebase.

When to use Gemini 3.8 Flash

Choose it when you need high-volume multimodal input, long context, structured tool use or a cost-efficient model for coding agents. It is particularly attractive before the introductory pricing expires.

Do not select it solely from a benchmark headline. Measure cost per successful task, tool-call accuracy, output-token use, latency and human review time. Computer use should remain behind preview-appropriate controls and monitoring.

Verdict

Gemini 3.8 Flash is production-ready as a model, with unusually broad input modalities and strong agent tooling at an introductory Flash price. The two planning constraints are equally important: its migration rules can break older request payloads, and its paid token rates double on January 1, 2027.