Overview
Gemini 3.5 Flash is Google DeepMind's latest Flash model, in preview and pitched as "frontier performance across agents and coding" at Flash-level latency and cost. It keeps Gemini's 1-million-token input window and 64K output, and is natively multimodal across text, image, video, audio and PDF inputs — with function calling, structured output, search-as-a-tool and code execution built in.
Where Gemini 3 Flash was a fast, cheaper sibling to the Pro line, 3.5 Flash narrows the intelligence gap dramatically: Google's own charts show it beating Gemini 3.1 Pro on a range of coding, agentic and multimodal benchmarks (76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas) while staying in the Flash price band. Armadin reports roughly 42% better results than Flash 3 on a long-range cyber benchmark at 68% better token efficiency, making it a strong default for high-volume agentic and coding work.
Key capabilities
- 1M-token context, 64K-token output
- State-of-the-art agentic coding for the Flash tier
- Truly multimodal — text, image, video, audio, PDF
- Beats Gemini 3.1 Pro on several coding and agent benchmarks
At a glance
Terminal-Bench 2.1
76.2%
SWE-Bench Pro
55.1%
MCP Atlas
83.6%
Context
1M tokens
Pricing: Low-cost Flash tier — built for high-volume, latency-sensitive work.
Pros & cons
What we like
- Near-Pro intelligence at Flash pricing
- Excellent agentic coding and multimodal reasoning
- Huge 1M context with native tool use
- Strong token-efficiency gains over Flash 3
Trade-offs
- Still in preview — rough edges expected
- Slightly behind the absolute frontier on the hardest reasoning
The verdict
Gemini 3.5 Flash is the model to default to when you want frontier-adjacent intelligence at Flash speed and cost. It closes most of the gap to the Pro line on coding and agentic work, making it the smart pick for high-volume production and real-time agent loops.
Best for: High-volume agentic coding and multimodal work at Flash-tier cost.
Frequently asked questions
How is Gemini 3.5 Flash different from 3 Flash?
It's a generational leap, not a tweak — Google reports big gains across coding, agentic and multimodal benchmarks (76.2% on Terminal-Bench 2.1 vs 58.0% for Flash 3), markedly better token efficiency, and it even beats Gemini 3.1 Pro on several measures while staying in the Flash price band.
Is Gemini 3.5 Flash multimodal?
Yes — it handles text, image, video, audio and PDF inputs natively, with function calling, structured output, search-as-a-tool and code execution available as built-in tools.