Updated

Gemini 3.5 Flash
Google's I/O 2026 Flash for agents and coding, launched ahead of Gemini 3.1 Pro on two agentic benchmarks.
What is Gemini 3.5 Flash?
Gemini 3.5 Flash is the Flash model Google released at Google I/O on May 19, 2026.
Google called it its strongest agentic and coding model at launch, ahead of Gemini 3.1 Pro on Terminal-Bench 2.1 and MCP Atlas, and said it ran four times faster than other frontier models in output tokens per second. Super Agent runs it on Google Cloud's Vertex AI, which lists its retirement for May 19, 2027 or later.
- Provider
- Context window
- 1M tokens
- Input
- Text, Image, Video, Audio, PDF
- Reasoning
- Low, medium, high
- Max output
- 65,536 tokens
- Knowledge cutoff
- January 2025
- Released
- May 19, 2026
How Gemini 3.5 Flash performs
Published by Google
Results Google reported for Gemini 3.5 Flash. Each lab runs its own setup, so compare these with other models only inside the same table.
| Benchmark | Score | Setting |
|---|---|---|
| Terminal-Bench 2.1 | 76.2% | Terminus-2 harness |
| SWE-Bench Pro (Public) | 55.1% | single attempt |
| MCP Atlas | 83.6% | |
| Toolathlon | 56.5% | |
| OSWorld-Verified | 78.4% | |
| Finance Agent v2 | 57.9% | |
| CharXiv Reasoning | 84.2% | no tools |
| MMMU-Pro | 83.6% | no tools |
| Humanity's Last Exam | 40.2% | |
| ARC-AGI-2 | 72.1% |
Source: deepmind.google
On LMArena
People compare two anonymous answers and vote for the better one. Ranked here against the other chat models you can pick in Super Agent.
Text
#14 of 27 in Super Agent
All text prompts, with style control on. #30 of 418 on the full leaderboard, from 48,504 votes.
- 1Claude Opus 5.51509
- 2Claude Fable 51503
- 3Claude Fable 5.11503
- 4Claude Opus 51491
- 5Gemini 3.7 Flash1490
- 14Gemini 3.5 Flash1477
Coding
#21 of 27 in Super Agent
Text prompts that ask for code. #62 of 413 on the full leaderboard, from 13,851 votes.
- 1Claude Fable 51551
- 2Claude Opus 5.51548
- 3GPT-6 Astra1543
- 4Kimi K31539
- 5Claude Fable 5.11538
- 21Gemini 3.5 Flash1508
WebDev
#24 of 27 in Super Agent
Building working web apps in Code Arena. #62 of 149 on the full leaderboard, from 9,382 votes.
- 1Claude Opus 5.51807
- 2GPT-6 Astra1775
- 3Claude Sonnet 5.51768
- 4GPT-6.1 Sol1749
- 5Claude Fable 5.11743
- 24Gemini 3.5 Flash1499
Math
#11 of 27 in Super Agent
Text prompts with math problems. #21 of 404 on the full leaderboard, from 2,391 votes.
- 1Claude Fable 5.11529
- 2Claude Fable 51521
- 3Claude Opus 51517
- 4GPT-6.1 Sol1513
- 5Claude Opus 5.51512
- 11Gemini 3.5 Flash1497
Hard prompts
#15 of 27 in Super Agent
Complex, multi-part and demanding text prompts. #43 of 418 on the full leaderboard, from 32,399 votes.
- 1Claude Opus 5.51535
- 2Claude Fable 51529
- 3Claude Fable 5.11523
- 4Kimi K31516
- 5Claude Opus 51515
- 15Gemini 3.5 Flash1495
Creative writing
#11 of 27 in Super Agent
Stories, poems and other original writing. #24 of 416 on the full leaderboard, from 9,477 votes.
- 1Claude Opus 5.51520
- 2Claude Fable 51501
- 3Gemini 3.7 Flash1492
- 4Claude Fable 5.11487
- 5Gemini 3.1 Pro Preview1481
- 11Gemini 3.5 Flash1464
Vision
#10 of 23 in Super Agent
Prompts that include images, with style control on. #24 of 167 on the full leaderboard, from 13,862 votes.
- 1Claude Fable 51309
- 2Gemini 3.7 Flash1295
- 3GPT-6.1 Sol1294
- 4Claude Opus 5.51293
- 5Claude Opus 51288
- 10Gemini 3.5 Flash1282
Source: LMArena, leaderboard published . Scores move as new votes come in.
What you can do with Gemini 3.5 Flash
Ahead of Gemini 3.1 Pro on Terminal-Bench 2.1 and MCP Atlas at launch
Four times faster output than other frontier models, Google said
Built for long-horizon agent tasks and subagents
Takes PDFs, audio and video as well as text and images
How to prompt Gemini 3.5 Flash
- Attach a PDF brief and ask for a summary or a shot list from it.
- Use low effort for quick replies; it is built for speed.
- Split big jobs into separate tasks that can run side by side.
Where Gemini 3.5 Flash fits best
Parallel tasks in an agent run
Interactive web UI ideas
Reading PDFs and long documents
Everyday multimodal chat
How Gemini 3.5 Flash compares
| Model | LMArena text | LMArena coding | Context window | Reasoning |
|---|---|---|---|---|
| Gemini 3.5 Flash | 1477#30 | 1508#62 | 1M tokens | Low, medium, high |
| Claude Sonnet 5.5 | 1478#29 | 1528#26 | 1M tokens | Low to max |
| GPT-6 Astra | 1477#32 | 1543#9 | 1.05M tokens | Low to max |
| GPT-6.1 Sol | 1480#27 | 1537#14 | 1.05M tokens | Low to max |
| GLM-5.3 | 1481#26 | 1521#33 | 1M tokens | Low, medium, high, max |
Good to know before you start
- Gemini 3.6 Flash and 3.7 Flash are newer Flash models in the same picker
- Output is text only, even though it accepts audio and video; media comes from Flashloop's generation models
Start chatting with Gemini 3.5 Flash
Chat with Gemini 3.5 Flash in Super AgentSelect Gemini 3.5 Flash
Open Super Agent on Flashloop and choose Gemini 3.5 Flash in the model picker.
Enter your prompt
Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.
Send and keep going
Send the message. Gemini 3.5 Flash answers in Super Agent and runs any generations you asked for.
Questions about Gemini 3.5 Flash
Which model answers when I pick Gemini 3.5 Flash?
Gemini 3.5 Flash itself. Super Agent calls it through Google Cloud's Vertex AI. On October 8, 2026 Google started sending requests for this name to Gemini 3.6 Flash on the Gemini API, but Vertex AI still serves Gemini 3.5 Flash and lists its retirement for May 19, 2027 or later.
What is Gemini 3.5 Flash good at?
Agentic and coding work. At launch Google said it beat Gemini 3.1 Pro on Terminal-Bench 2.1 and MCP Atlas, and it became the default model in the Gemini app.
What inputs does Gemini 3.5 Flash accept?
It takes text, images, video, audio and PDFs within a 1M-token window.
How is Gemini 3.5 Flash different from Gemini 3.5 Flash Lite?
Flash Lite is Google's fastest, lowest-cost 3.5-class model for high-volume work and subagent tasks. Gemini 3.5 Flash is the fuller model for coding and agents.
More chat models on Flashloop
Claude Sonnet 5.5Anthropic's upgrade to Sonnet 5 for everyday building, bug fixing and polished documents.
GPT-6 AstraOpenAI's GPT-6 flagship, with top scores on computer use, browsing and software engineering.
GPT-6.1 SolOpenAI's GPT-6 Sol upgrade, close to GPT-6 Astra on agentic coding and computer use for far less.
GLM-5.3Z.ai's open-weight, text-only model for long coding runs and security work, with 1M context.