Gemini 3.8 Flash
Gemini 3.8 Flash is the model Google calls its "most intelligent Flash model", engineered for long-horizon software engineering, autonomous agents and complex enterprise workflows.
Specifications & pricing.
- FIELD
- DOCUMENTED VALUE
- SOURCE
- Input modalities
Text, image, video, audio and PDF
Model page
- Output modality
Text
Model page
- Input token limit
1,048,576
Model page
- Output token limit
65,536
Model page
- Supported
Caching, code execution, file search, function calling, grounding with Google Maps, Search grounding, structured outputs, URL context, thinking (low, medium, high)
Model page
- Preview only
Computer use
Model page
- Not supported
Audio generation, image generation, Live API; the
minimalthinking level returns an errorModel page
- Consumption options
Batch API, Flex inference and Priority inference all supported
Model page
- Versions
Stable:
gemini-3.8-flash; tagged "New Stable" in Google's listModel page; Gemini models list
- Latest update
September 2026
Model page
- Standard price, paid tier
USD 0.75 input / USD 3.75 output per 1M tokens (output includes thinking tokens) through 31 December 2026; USD 1.50 / USD 7.50 from 1 January 2027, as published on 15 October 2026
- Context caching
USD 0.075 per 1M tokens (USD 0.15 from 2027) plus hourly storage
Pricing page
- Batch, Flex, Priority
Batch and Flex USD 0.375 / USD 1.875; Priority USD 1.35 / USD 6.75 per 1M tokens through 2026, each doubling from 2027
Pricing page
When to use it
What Google says. The model page pitches 3.8 Flash at long-horizon software engineering, autonomous agents and enterprise workflows "with the speed and cost efficiency of Flash". The models list describes Gemini 3.6 Flash as the "previous-generation Flash model", which makes 3.8 Flash the current one. The same page tells new projects to use 3.5 Flash-Lite or 3.8 Flash.
Our editorial judgment. The documented feature set reads like a model built for agents: function calling, code execution, file search, URL context, grounding and preview computer use, with three thinking levels to trade cost against depth. That makes it a sensible first candidate for a coding or research agent that needs more headroom than a Lite model. For chat or extraction at high volume, start with the cheaper sibling and move up only if quality fails. If you are choosing a coding assistant rather than a raw API model, see our AI coding tools guide.
Limits and costs to plan for
- Scheduled price rise. Every paid token rate for this model doubles on 1 January 2027.
- Thinking tokens are billed as output. Higher thinking levels raise cost per call.
- No
minimalthinking. It returns an error; uselow. - Text out only. Audio generation, image generation and the Live API are not supported; Google lists separate models for those jobs.
- Computer use is preview. Expect changes before building production flows on it.
- 65,536 output cap. Long generations may need chunking.
- Free tier data use. Google marks free tier traffic as used to improve its products; paid traffic is not.
- Grounding is metered. Search and Maps grounding include 5,000 free requests per month shared across Gemini 3 and newer models.
How it compares within the family
Both models share the same documented shape: five input types, text output, 1,048,576 input and 65,536 output tokens, and the same tool list. The differences are price, thinking options and positioning.
| Model | Standard input / output (USD per 1M tokens) | Thinking levels documented | Latest update |
|---|---|---|---|
| Gemini 3.8 Flash | 0.75 / 3.75 through 2026; 1.50 / 7.50 from 2027 | low, medium, high | September 2026 |
| Gemini 3.5 Flash-Lite | 0.30 / 2.50 | "Supported", no levels listed | July 2026 |
Google describes 3.5 Flash-Lite as its fastest, most cost-effective 3.5 model for high-throughput execution. At 2027 rates, 3.8 Flash costs five times as much per input token and three times as much per output token. Outside Google, DeepSeek's docs list DeepSeek V4.1 Flash at USD 0.30 input (cache miss) and USD 1.20 output per 1M tokens at peak, and DeepSeek V4 Pro at USD 1.32 and USD 3.96.
How to evaluate it on your work
- Collect 20 to 50 real tasks with known good answers, weighted toward agentic or coding work.
- Run each task on 3.8 Flash at low and at high thinking, and on 3.5 Flash-Lite, with identical prompts and tools.
- Record pass or fail and token counts, then price each run at 2026 and 2027 list rates.
- Rerun the cheapest passing setup through Batch or Flex if your work can wait, since those rates are half of Standard.
- Keep 3.8 Flash only where its pass rate covers the 2027 price gap.
Method and limits
We read the Gemini 3.8 Flash model page, the Gemini models list and the Gemini API pricing page on 15 October 2026. We have not run Gemini 3.8 Flash, and nothing here is a test result or ranking. Prices and features change; check Google's pages before committing budget. See our ranking method for how HumanRank scores tools and the model directory for other profiles.
Frequently asked questions
What is Gemini 3.8 Flash? It is Google's current Flash model, described as its most intelligent Flash model for long-horizon software engineering, autonomous agents and enterprise workflows.
How much does Gemini 3.8 Flash cost? It costs USD 0.75 per 1M input tokens and USD 3.75 per 1M output tokens on the paid Standard tier through 31 December 2026, then USD 1.50 and USD 7.50 from 1 January 2027.
Does Gemini 3.8 Flash support thinking?
Yes, at low, medium and high levels; the minimal level is not supported and returns an error.
What is the Gemini 3.8 Flash context window? It is 1,048,576 input tokens, with a 65,536 token output limit.
Can Gemini 3.8 Flash generate images or audio? No, it outputs text only, and Google lists image generation, audio generation and the Live API as not supported.
Is Gemini 3.8 Flash worth it over Gemini 3.5 Flash-Lite? Only if your tests show it completes more of your tasks, since Flash-Lite is cheaper on input and output.