Gemini 3.1 Flash-Lite

2026-03-03 · Publisher : Google · Domain : Artificial Intelligence

Google launches Gemini 3.1 Flash-Lite, its fastest and cheapest generation 3 model, at USD 0.25 per million input tokens, for translation, classification and real-time use.

What changes

  • Optimized for high-volume, low-latency tasks (translation, classification).
  • Throughput of 363 tokens per second announced in the model card.
  • Clear gain over Gemini 2.5 Flash-Lite in reasoning, multimodal and code.
  • Model based on Gemini 3 Pro according to its card.

### Specifications - Context: up to 1 million tokens; output up to 64,000 tokens. - Modalities: text, images, audio and video as input, text as output. - Availability: Gemini API (AI Studio) and Vertex AI; GA version on May 7, 2026. - API identifier: gemini-3.1-flash-lite (preview: gemini-3.1-flash-lite-preview). ### Pricing | Item | Price (USD / 1M tokens) | |---|---| | Input (text, image, video) | 0.25 | | Audio input | 0.50 | | Output | 1.50 | Batch: 0.125 for input and 0.75 for output (current pricing page). ### Benchmarks | Benchmark (scores published by the vendor) | 3.1 Flash-Lite | 2.5 Flash | GPT-5 mini | |---|---|---|---| | GPQA Diamond | 86.9% | 82.8% | 82.3% | | MMMU-Pro | 76.8% | 66.7% | 74.1% | | Humanity's Last Exam (no tools) | 16.0% | 11.0% | 16.7% | | LiveCodeBench | 72.0% | 62.6% | 80.4% | | MMMLU | 88.9% | 86.6% | 84.9% | ### Key takeaways A low-cost tier for mass processing (classification, translation, extraction), to be compared with 2.5 Flash-Lite, which remains cheaper but is markedly less capable.

Links

Tags : LLM, Google, Gemini, Flash-Lite, Faible coût