Pro-model performance at the price and speed of a Flash model. Multimodal reasoning, a 1M-token context window and agent-ready capabilities.
Gemini 3 Flash delivers reasoning quality close to Pro models, but at a fraction of the latency. It outperforms Gemini 2.5 Pro on efficiency, enabling real-time workflows.
Thinking Levels (Auto-Config)
Tokens of Context
Upload entire books, hundreds of pages of documentation or hours of video footage. The model remembers everything.
This isn't a text model with add-ons. Gemini 3 Flash natively "sees" images, "hears" audio and analyses video in a single chain of thought.
Built to get tasks done. It supports Structured Outputs and Tool Use with high reliability (90%+ in benchmarks).
Reliability Score: 90.2%
Official benchmark results (Late 2025). Compared against Gemini 3 Pro, Claude Sonnet 4.5, GPT-5.2 and Grok 4.1 Fast.
| Benchmark | Gemini 3 Flash | Gemini 3 Pro | GPT-5.2 | Claude 4.5 | Grok 4.1 |
|---|---|---|---|---|---|
| Math (AIME 2025) | 95.20% | 95.00% | 100% | 87.00% | 91.90% |
| Science (GPQA) | 90.40% | 91.90% | 92.40% | 83.40% | 84.30% |
| MMMU-Pro | 81.00% | 81.20% | 79.50% | 68.00% | 63.00% |
| Video-MMMU | 86.90% | 87.60% | 85.90% | 77.80% | — |
| Coding (LiveCode) | 2316 | 2439 | 2393 | 1418 | — |
| Agentic (τ²-bench) | 90.20% | 90.70% | — | 77.80% | — |
| Context (128k) | 67.20% | 77.00% | 81.90% | 47.10% | 54.60% |
Interactive visual reasoning. Upload complex charts, multi-page PDFs (up to 1,000 pages) or technical documentation. Gemini 3 Flash will extract the key insights in real time.
Turn napkin sketches or UI mockups into working HTML/CSS code. The model understands the visual layout and converts it into structure.
Extract metadata from video, summarise long meetings, analyse sports footage frame by frame. All in near real time.
Generate content tailored to each user's context from multiple sources (history, files, preferences) in real time.
Thanks to the 1M window, you can paste an entire code repository and the model will find logic bugs and suggest optimisations.
Build autonomous systems that plan tasks, use external tools and carry out multi-step operations without supervision.
Signing up takes only a moment. You get a starter pack to try the Frontier model for free.
Use the ready-made chat or the Studio module for multimodal tasks (image/video/code).
Drop in a PDF, video or paste your code. Set the Reasoning level and get your result in seconds.
See the power of Gemini 3 Flash in action