Gemini 3.5 Flash-Lite
Google's fastest, lowest-cost Gemini 3.5-class model for high-volume agent and document work
Context window
1M
Input / 1M tokens
$0.30
Output / 1M tokens
$2.50
Provider
Gemini API paid tier, per 1M tokens: input $0.30 for text, image, video and audio alike; output $2.50 including thinking tokens; cached input $0.03 plus $1.00 per 1M tokens per hour of storage. Batch is half price ($0.15/$1.25). No introductory pricing applies. Max output is 65,536 tokens. · Data verified 2026-11-11
Gemini 3.5 Flash-Lite (July 21, 2026) is Google's fastest and most cost-effective model in the Gemini 3.5 class. Built on Gemini 3.1 Flash-Lite, it targets high-throughput, latency-sensitive work such as subagent tasks, agentic search, translation, classification and document parsing. It accepts text, image, video, audio and PDF input with a 1,048,576-token context window and up to 65,536 output tokens, and developers can pick minimal, low or higher thinking levels. Computer use is a built-in tool. Google's model card shows large gains over 3.1 Flash-Lite, such as 54.2% vs 38.3% on SWE-bench Pro, at a higher price ($0.30/$2.50 vs $0.25/$1.50 per 1M tokens). From December 1, 2026, Gemini API requests to the deprecated gemini-2.5-flash are routed to this model.
How to access it
Gemini API through Google AI Studio (model ID gemini-3.5-flash-lite, listed as stable) and Android Studio; enterprises use it through Gemini Enterprise Agent Platform. Google's launch post says it is also rolling out in Google Search.
Strengths
- ✓Low price for its class: $0.30 input / $2.50 output per 1M tokens, half that on Batch
- ✓1M-token multimodal context (text, image, video, audio, PDF)
- ✓Adjustable thinking levels from minimal to high
- ✓Built-in computer use, Google Search and Maps grounding, code execution and function calling
- ✓About 350 output tokens per second (Artificial Analysis figure cited by Google)
Best for developers who...
When to choose it (and when not to)
Reach for Gemini 3.5 Flash-Lite when...
- →You run large volumes of short or structured tasks where cost per call matters
- →You need a fast subagent or router under a larger model such as Gemini 3.8 Flash
- →You are moving off gemini-2.5-flash before Google reroutes it on December 1, 2026
Look elsewhere if...
- ✕Hard multi-step coding or reasoning (Gemini 3.8 Flash or Gemini 4 Argon score higher)
- ✕You need the lowest possible token price: Gemini 3.1 Flash-Lite costs $0.25/$1.50 per 1M tokens
- ✕Real-time voice (it does not support the Live API)
How to use it
- ›Set the thinking level to minimal for simple high-volume calls and raise it for multi-step subagent work
- ›Use the Batch API for non-urgent jobs to halve the cost
- ›Ask for structured outputs with a JSON schema when extracting fields from documents
- ›Cache long shared prefixes such as instructions or reference documents
Quickstart
Pythonfrom google import genai
client = genai.Client()
interaction = client.interactions.create(
model="gemini-3.5-flash-lite",
input="Classify this support ticket as billing, bug or other: ...",
)
print(interaction.output_text)Install with pip install -U google-genai. Pattern from Google's Gemini API quickstart (Interactions API).
API model id: gemini-3.5-flash-lite
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Pro | 54.2% | SWE-Bench Pro (Public) in Google's model card; Gemini 3.1 Flash-Lite 38.3%, GPT-5.4 mini 54.4% |
| Terminal-Bench 2.1 | 54.0% | Google model card, Terminus-2 harness; Gemini 3.1 Flash-Lite 31.0% |
| OSWorld-Verified | 74.0% | Google model card; Gemini 3.1 Flash-Lite 54.3% |
| GDPval-AA v2 | 1140 Elo | Google model card; Gemini 3.1 Flash-Lite 642 |
| MLE-Bench | 39.2% | Google model card; Gemini 3.1 Flash-Lite 22.0% |
| GDM-MRCR v2, 8-needle 128K average | 72.2% | Google model card long-context test; Gemini 3.1 Flash-Lite 60.1% |
Source: Gemini 3.5 Flash-Lite model card
Compare Gemini 3.5 Flash-Lite
Compare Gemini 3.5 Flash-Lite with any other model
Build a comparison →All model comparisons →