Strengths
What it's the best tool for
- Pro-grade reasoning at Flash latency
- Adjustable thinking level per task
- 30% fewer tokens than Gemini 2.5 Pro
- GPQA Diamond 90.4%, MMMU Pro 81.2%
- Multimodal, documents out of the box
Limitations
When to reach for something else
- Smaller context than Pro variants
- No native image generation
- High thinking level increases latency
- Strict content filters
Sample output
How Gemini 3 Flash responds
Prompt
Read this 40-page PDF and list 10 key findings with quoted evidence. Use thinking=medium, be concise.
Gemini 3 Flash
1) 2025 revenue up 34% YoY (p. 5).
2) EBITDA margin steady at 21% (p. 9).
3) New markets: Turkey, Brazil (p. 14).
Remaining 7 findings with citations below.
Where teams use it
Four scenarios where it pays for itself
01
Production chat
Frontier replies, cheap
02
PDF analysis
Thinking scaled to task
03
Agents
Fast tool-use loops
04
Classification
High-volume batches
About model
More about Gemini 3 Flash
Gemini 3 Flash Online — Google's Fast Frontier LLM
Gemini 3 Flash shipped on December 17, 2025 as Google's speed-tier Gemini 3: Pro-grade reasoning at Flash speed. Live on NetRoom with no VPN.
Under the hood
A thinking level parameter (minimal/low/medium/high) controls reasoning depth. media_resolution manages visual tokenisation. Burns 30% fewer tokens than Gemini 2.5 Pro at the same accuracy.
Benchmarks
GPQA Diamond 90.4%, Humanity's Last Exam 33.7% without tools, MMMU Pro 81.2% — on par with Gemini 3 Pro.
Where it fits
Production chat, agent loops, batch document work, classification — any workload balancing price and quality.
Try Gemini 3 Flash
right now
Free access to basic models. No card, no obligations.