Gemini 3 Flash Preview is a high speed, high value thinking model designed for agentic workflows, multi turn chat, and coding assistance. It delivers near Pro level reasoning and tool use performance with substantially lower latency than larger Gemini variants, making it well suited for interactive development, long running agent loops, and collaborative coding tasks. Compared to Gemini 2.5 Flash, it provides broad quality improvements across reasoning, multimodal understanding, and reliability.
The model supports a 1M token context window and multimodal inputs including text, images, audio, video, and PDFs, with text output. It includes configurable reasoning via thinking levels (minimal, low, medium, high), structured output, tool use, and automatic context caching. Gemini 3 Flash Preview is optimized for users who want strong reasoning and agentic behavior without the cost or latency of full scale frontier models.
Evaluations
8
across 8 benchmarks
Latency
629ms
Context length
1M
tokens
Cost
$0.50 · $3
input · output per 1M tokens
Key takeaways
The Gemini 3 Flash Preview (minimal) is a Transformer-based model optimized for agentic workflows, multi-turn chat, and coding assistance, excelling in strong reasoning and agentic behavior without the high cost or latency of larger models.
The model supports multimodal inputs (text, images, audio, video, PDFs) and generates text output, featuring configurable reasoning levels, structured output, tool use, and automatic context caching.
It demonstrates robust mathematical reasoning, achieving 95% accuracy on MATH-500, particularly strong in algebra and number theory, and adept at applying mathematical theorems and handling complex number arithmetic.
Key limitations include inconsistent adherence to numerical values/diagrams, struggles with interpreting specific geometric relationships, and occasional errors in algebraic simplification and multi-step numerical calculations, leading to failures in complex code integrations.
In coding tasks (SWE-bench Lite), the model achieves 12.67% accuracy, showing proficiency in debugging numerical computation libraries (Astropy) and certain Django features (HttpResponse, MigrationRecorder), but struggles with complex Django ORM queries, inheritance handling, and accurate serialization of complex Python objects.
