Gemini 3.5 Flash Lite
GeminiCompact
ThinkingTool UseVisionStructured Output
About this model
Gemini 3.5 Flash Lite is a high-efficiency model from Google with upgraded agentic capabilities. It is suited for subagents that execute focused tasks within complex, multi-agent workflows.
Performance Tier
Compact
Gemini 3.5 Flash Lite is a compact model from Gemini : optimized for speed and affordability.
Small, fast, and affordable. Optimized for speed and low cost, great for high-volume or simple tasks.
Pricing
This model is included in Elosia plansAffordable
Low cost. Suitable for sustained use and high-volume interactions.
| Type | per 1M tokens |
|---|---|
| Input (prompt) | $0.300 |
| Output (completion) | $2.50 |
| Image | $0.300 |
| Internal reasoning | $2.50 |
| Cache read | $0.030 |
| Cache write | $0.083 |
Capabilities
Context Length1.0M
Max Output Tokens66K
TokenizerGemini
Inputtext, image, video, file, audio
Outputtext
Release DateJuly 21, 2026
Benchmarks
General Intelligence
MMLU
Not reported
GPQA Diamond
83.8%
Mathematics
MATH-500
Not reported
Programming
HumanEval
Not reported
Reasoning
Humanity's Last Exam
17.5%
Multimodal
MMMU-Pro
79%
Agentic
SWE-bench Pro
54.2%
Terminal-Bench 2.1
54%
Where does Gemini 3.5 Flash Lite stand?
Compare its performance index against every other model.
Recommended Use Cases
General ChatSummarizationData ExtractionCustomer SupportTranslation
Strengths
- Ultra-low-cost, high-speed inference (~448 tokens/s) at $0.30/$2.50 per M tokens
- Outperforms the larger Gemini 3 Flash on agentic coding (SWE-bench Pro 54.2% vs 49.6%)
- Strong knowledge and multimodal grounding for a lite model (GPQA Diamond 83.8%, MMMU-Pro 79.0%)
- 1M-token context with full multimodal input (text, image, video, audio) and a built-in thinking mode
Limitations
- Weak on the hardest reasoning (Humanity's Last Exam 17.5%): a lite model, not a frontier reasoner
- Google reports only agentic and coding evals; the GPQA, MMMU-Pro and HLE scores here come from third-party testing (Artificial Analysis)
Frequently asked questions
Resources
This model may use your data for training
