Gemini 3.5 Flash-Lite

// Overview

Gemini 3.5 Flash-Lite is Google's cheapest and fastest Gemini 3 model, released July 21, 2026 as the API model gemini-3.5-flash-lite. It sits under Gemini 3.6 Flash in the Gemini 3 hierarchy and is aimed at high-volume, latency-sensitive workloads — translation, classification, tagging, large-scale summarisation, and cost-sensitive agent loops.

Despite being the low end of the line, Google reports Gemini 3.5 Flash-Lite outperforms the older, larger Gemini 3 Flash on several agentic and coding evals: 54.2% vs 49.6% on SWE-Bench Pro, 74.0% vs 65.1% on OSWorld-Verified, and 54% vs 31% (Gemini 3.1 Flash-Lite) on Terminal-Bench 2.1. Artificial Analysis measured its speed at about 350 output tokens/second and its intelligence-index score at 36, both well above the median for reasoning models in its price tier.

The model accepts text, images, audio, video, and PDFs as input and returns text, with a 1 million-token context window and up to 64K output tokens. It is available through the Gemini API, Google AI Studio, Vertex AI, the Gemini app, and Search AI Overviews at $0.30 per million input tokens and $2.50 per million output tokens (with cached input at $0.03 per million). Its knowledge cutoff is March 2026.

// Benchmarks

This model's scores

1. SWE-Bench Pro54.2%
2. OSWorld-Verified (agentic computer use)74%
3. Terminal-Bench 2.1 (agentic terminal coding)54%
4. GDM-MRCR v2 (long-context retrieval)72.2%
5. GDPval-AA v2 (economically valuable knowledge work)1140Elo
6. Artificial Analysis Intelligence Index36

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Standard paid tier; output includes thinking tokens. Cached input storage $1.00 per 1M tokens per hour. A free tier with usage limits is also available.

Pricing source ↗

// Strengths

• Very low price: $0.30 input / $2.50 output per million tokens, with cached input at $0.03 per million
• About 350 output tokens/second on the Artificial Analysis Index — well above the median for reasoning models in its price tier (~107 t/s)
• Beats the older Gemini 3 Flash on SWE-Bench Pro (54.2% vs 49.6%) and OSWorld-Verified (74.0% vs 65.1%)
• Terminal-Bench 2.1 rises to 54% (vs 31% for Gemini 3.1 Flash-Lite) — Flash-Lite is now usable for terminal-style agents
• 1M-token multimodal context (text, image, audio, video, PDF) with 64K output

// Best for

• Reach for it as the default cheap tier for high-volume translation, classification, and tagging
• Reach for it for cost-sensitive agent loops where you would previously have picked a Flash-tier model
• Reach for it when you need latency below what larger Flash tiers offer
• Reach for it for multimodal ingestion at scale — audio/video/images/PDFs cheaply within a 1M-token window

// How to access

// Gemini Flash-Lite — every version

The full lineage of the Gemini Flash-Lite line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.