Skip to content
AIMODELRACE
Google DeepMindLanguage model

Gemini 3.5 Flash

Absurdly fast frontier-lite model with a free tier and strong agentic chops.

Compare head-to-headRace it in 3D

Overall Intelligence

#16

of 56
6.6 pts behind #1

3

races led

60

weeks at #1

82.2

score

Released
May 19, 2026
5 mo ago
Context
1M
65K max output
Price
$0.3 / $2.5
per 1M in/out · $0.85 blended
Speed
524 tok/s
0.38s to first token
Modalities
text · image · audio · video
outputs text
Weights
Closed
API and apps

The 30-second version

An absurdly fast frontier-lite model with a generous free tier. Gemini 3.5 Flash streams over 500 tokens per second and is surprisingly capable at agentic tasks.

Key takeaways
  • 1Leads our speed race by a dominant margin and has the lowest latency.
  • 2Strong multimodal input and agentic skills for a cheap model.
  • 3$0.30 / $2.50 per million tokens with a free tier for experimentation.
  • 4A great default for real-time features and high-volume pipelines.

What you need to know

Purpose

Speed and scale: real-time assistants, voice, high-volume extraction and agents that need many fast steps.

What Google optimised for

Throughput and latency on Google's TPU stack, keeping multimodal input and tool use strong.

Unique strengths

  • Over 500 tokens per second.
  • Sub-half-second first token.
  • Multimodal input at commodity prices.
Insight
How it differs from rivals. Faster than every rival by a wide margin. Versus Haiku 5.5: faster and better with images and video, weaker at computer use.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

CodingHard ReasoningMathScienceWriting & LanguageMultimodalTool Use & AgentsLong ContextTrust & FactualitySpeed & Cost

3.5 FlashBest on each axis

Overall Intelligence over time

#1 frontier3.5 FlashKimi K3Spark Open
55.060.065.070.075.080.085.090.02025Apr ’25Jul ’25Oct ’252026Apr ’26Jul ’26Oct ’26

Best use cases

01

Real-time UX

Autocomplete, live summarisation and voice assistants.

02

High-volume pipelines

Classifying, extracting and tagging millions of items.

03

Fast agent steps

Loops with many small tool calls where latency compounds.

Strengths

  • Speed

    Dominant lead in our output-speed race.

  • Latency

    Lowest time to first token in the race.

Weaknesses

  • Depth

    Clearly behind frontier models on hard reasoning and long tasks.

When to choose it, and when not to

3.5 FlashoverGPT-6 Luna

Speed and image or video input matter most.

Pick something else when…

  • Complex, high-stakes reasoning.
  • Long-form writing quality.

Update history

The distilled version first; open any release for the deeper breakdown.

The Gemini Flash family

2.0 FlashFebruary 20253 FlashDecember 20253.5 FlashMay 2026