Google
Google

Gemini 3.5 Flash

Release Date 2026-05-19(Knowledge cutoff 2025-01)10.5M IN · 66K OUTProprietary ModelAPI available

Gemini Flash is Google's multimodal reasoning model for text generation, coding and tool-connected workflows. It targets Flash-class efficiency with coding and parallel agent work, while search, code execution and function calling depend on the configured environment rather than automatic control of every app. AIB places coding in the middle of the pack, a current comparative position rather than a guaranteed success rate. Long inputs and tool access do not guarantee complete understanding or correct actions, so verify source-grounded answers and tool results on real tasks. Compare response time and total token use with correction effort, retries and service terms on representative inputs, rather than choosing solely from a demonstration or a headline benchmark.

API cost

Official price
$1.5IN$9OUT
SourceGoogle official docs

Output Speed

Very fastper 2,000 tokens
150tok/s
First Output 1.6stotal: 15.0s
SourceOpenRouterLast updated

User Comparisons

Writing Style
Answer length
Average
top 44%
Vocabulary diversity
Average
bottom 46%
Sentence rhythm
Average
bottom 50%
Language mixing
Average
top 43%
List usage
Often
29% of lines
Heading usage
Occasionally
13% of lines
Reasoning share
Very heavy
53% of total
Lies (Hallucination)
Honest answers
75%
Correct 6Honest don't-know 3Lies 3
Lies without doubt
3 questions
Response Speed
instability ±40%
First response 7.9sTotal 14.5s
176 tok/s
5.2ms per token
User Reviews
  • •

    Fast and stable responses

  • •

    At least for translation, Gemini 3.5 Flash seems much more natural and better.

  • •

    Fast and accurate answers!

  • •

    The responses are really fast.

SourceAIB Inc.Last updated

Expert Performance Evaluation

147674.633.6

More about Arena Text

32.2%$5,39695.6%83.6%

More about τ³-Banking

74.3%146327.5%31.5%57.9%64.7%21.7%30.8%55.8%146264.914.0%2.5%

By effortMinimal 61.3%Medium 74.3%High 73.3%

More about Long Context Reasoning

1478Reasoning effort: High66.2%5.7%

More about Arena Text Factuality

6.6%150649.091155.1%78.7%46.2%70.1149053.9%79.3%37.4%62.6%149778.20.0%48.7%6.7%

More about Terminal-Bench 4.0

1472147976.3%1469146684.6107776.6%145775.6

More about Arena Writing, Literature & Language

42.7%13.1%148292.2%95.6%88.262.8%26.8%31.0%58.1%1497149082.039.0%14.6%92.5%72.1%76.7%25.4%45.0%32.0%94.7%47.2

By effortMinimal 24.1%Medium 41.3%High 42.7%

More about Humanity's Last Exam

130633.6%84.9%85.0%83.6%

More about Arena Vision

Specialized fields

1447145714961488148814824.8%95.0%70.1%41.4%73.5%62.6%64.3%52.9%95.4%95.8%95.3%93.0%95.5%81.2%80.1%82.3%96.0%91.3%97.8%87.5%134255932393.9%89.9%93.3%89.9%45.0%79.0%89.6%93.0%82.8%87.0%92.0%56.0%32.3%82.5%89.9%53.2%82.5%86.1%81.5%70.4%67.3%22.0%

More about Arena Korean

OddBench

20.0%Reasoning effort: Extra High92.6%33426756925714

By effortMinimal 19.0%Extra High 20.0%

More about BullshitBench v2

See each evaluation’s detail page for its source and last updateData corrections & removal requests

Public Opinion

TL;DR

While users value Gemini 3.5 Flash for its impressive speed and multilingual translation, significant concerns persist regarding its reliability during complex coding tasks and data extraction from documents.

280 opinionsYouTube
  • 51%
    49%
    Positive
    Negative
  • YouTubeYouTube
    60%
    40%
    Positive
    Negative

Reaction Summary

1.

Operational Speed and Efficiency

39 opinions

Users frequently praise the model for its low latency and rapid response times, making it a preferred choice for quick, non-complex interactions.

2.

Multilingual Translation and Voice Features

39 opinions

Global users express excitement over real-time translation and voice capabilities, though some report issues with specific dialects and synchronization.

3.

Reliability and Hallucination Concerns

34 opinions

A significant portion of feedback highlights failures in deep research, data extraction from PDFs, and the invention of facts during complex tasks.

4.

Competitive Benchmarking and Coding Performance

32 opinions

Technical discussions often compare Flash unfavorably to Claude or Gemini Pro for coding, noting a gap in instruction following and reasoning.

Actual opinionsselected excerpts

  • RIP Fable 5 💔
    YouTube@ThatGuyGamer221176View original
  • Wait this might be tuff 🥹🔥🔥
    YouTube@tef-AI112View original
  • Tam geminiden ders çalışıyodum bildirim düştü😂
    YouTube@suleymanonderatılgan62View original
  • Gemini 3.5 Flash genuinely deserves more credit. I have tested it side by side with several top models using the same prompts, and it consistently surprises me. The multimodal understanding, visual reasoning, 3D comprehension, speed, and overall smoothness make it feel incredibly capable. Benchmarks are useful, but real world experience matters too, and Gemini 3.5 Flash has been one of the most impressive models I have used.
    YouTube@Adityasongs116View original
  • People in this thread underestimate Gemini. Of the 3 major proprietary models - Google, OpenAI and Anthropic - you can get Gemini to do a lot more things. Comparatively speaking its guardrails are sh!t. But that's a good thing in some cases..
    YouTube@smackyay16View original
  • Lo veo para una posibilidad que llevo años esperando, aunque aun es algo temprano, pero llegará... IMAGINA.... Ir al cine (las personas que vivimos en el extranjero) y ver una película en otro idioma y escuchar en tu idioma y con el mismo todo de voz la película. Creo que habre cientos de posibilidades... Respecto a lo de aprender un idioma, yo si soy partidario que se acabe ya esa frontera de tener que aprender otro idioma, el ser humano tiene potencial para dejar eso atrás... Imagina poder ir a cualquier país y entrar a un restaurante, hacer amistades, comprar.... sin necesidad de memorizar un extenso idioma... Buen video Crack...
    YouTube@titopeyo14View original
  • スプレッドシートにくっついたGeminiがだいぶ使いやすくなってきました。 関数全部お任せに
    YouTube@superlegend216View original
  • 3:20 şu özellik manyak bir özellik
    YouTube@MBurak715View original
  • AIも本物もどっちもめっちゃ可愛い❤
    YouTube@サイコバ-t5i12View original
  • Google is far from cooked..struck a nerve did we?!?!
    YouTube@Jovian72111View original
SourceAIB PulseLast analyzed