Gemini 3.8 Flash API & Pricing Guide

01

Gemini 3.8 Flash API quick facts

• Stable model ID: gemini-3.8-flash
• GA release: September 2, 2026
• Input limit: 1,048,576 tokens
• Maximum output: 65,536 tokens
• Inputs: text, image, video, audio and PDF
• Output: text
• Thinking levels: low, medium and high; medium is default

For the broader release context, capabilities and benchmark discussion, see the main Gemini 3.8 Flash overview.

02

API pricing through 2026 and from 2027

Google's introductory base price through December 31, 2026 is $0.75/M input tokens and $3.75/M output tokens.

From January 1, 2027, Google says standard pricing will be $1.50/M input and $7.50/M output. Recalculate production budgets before the pricing transition.

03

Batch, Flex and Priority inference

Google documents Gemini 3.8 Flash as supporting Batch, Flex and Priority inference. Flex is priced at 50% of the standard API rate and is intended for workloads that can tolerate a more flexible service profile.

Google's current inference documentation also describes Batch as a 50% discount path for asynchronous work with up to a 24-hour processing window. Priority is a premium low-latency tier and should be costed separately.

04

Start with a minimal request, then add tools

Use a current Google Gen AI SDK or REST request and specify gemini-3.8-flash as the model. Validate a plain prompt first, then add function calling, grounding, file search or computer use incrementally.

Google's migration guidance calls out thought-signature and tool-call behavior across Gemini 3 models. Keep SDKs current and preserve required tool metadata instead of copying assumptions from older integrations.

05

Thinking levels and cost control

Start with low thinking for latency-sensitive work, use medium as the general baseline and reserve high for tasks where deeper reasoning justifies extra latency and token use.

Measure total task economics. A higher reasoning spend can be worthwhile if it reduces retries, but it can also make a nominally cheap model expensive on long agent loops.

06

How to use the 1M-token context window

A 1,048,576-token input window can hold very large codebases, document sets and multimodal inputs, but filling the window on every request is rarely the cheapest architecture.

• Use retrieval when only a small part of a corpus is relevant
• Use context caching for repeated large prefixes
• Break long-running agent jobs into bounded steps
• Track token use per completed task rather than per request

07

Rate limits are tier-dependent

Google publishes Gemini API limits by account tier and says active limits can change with tier and account status. The current table lists Gemini 3.8 Flash across multiple tiers and separate Batch enqueued-token ceilings.

08

Production checklist

1. Use the stable gemini-3.8-flash model ID
2. Use a current Google Gen AI SDK
3. Benchmark low, medium and high thinking on representative tasks
4. Test tool calls individually before combining many tools
5. Cap retries and agent-loop depth
6. Require approval for consequential computer-use actions
7. Use caching or retrieval for repeated large context
8. Compare standard, Batch and Flex costs on real workloads
9. Model the January 2027 pricing increase
10. Recheck project-specific quotas before deployment

09

Gemini 3.8 Flash API FAQ

What is the exact model ID? gemini-3.8-flash.

Does it support Batch and Flex? Yes. Google's current docs list Gemini 3.8 Flash as supported for both, with Flex at 50% of standard API pricing.

Is computer use GA? No. The base model is GA, but Google still marks computer use as Preview.