Google Launches Gemini 3.1 Flash-Lite Preview via API

March 3, 2026

Details

  • Google introduced Gemini 3.1 Flash-Lite in preview, the fastest and most cost-efficient model in the Gemini 3 series, available now via Gemini API, Google AI Studio, and Vertex AI.\[1]\[3]
  • Priced at $0.25 per 1M input tokens and $1.50 per 1M output tokens, it offers enhanced performance over Gemini 2.5 Flash with 2.5x faster time to first token and 45% higher output speed.[1]
  • Features dynamic thinking levels for adjustable reasoning, supporting high-volume tasks like translation, content moderation, UI generation, simulations, and multimodal inputs (text, images, audio, video) with 1M token context.\[1]\[2][3]
  • Achieves Elo score of 1432 on Arena.ai, 86.9% on GPQA Diamond, 76.8% on MMMU Pro, outperforming similar-tier models and prior Gemini versions in reasoning and multimodal benchmarks.\[1]\[6]
  • Early adopters like Latitude, Cartwheel, and Whering praise its efficiency for complex, scalable workloads such as e-commerce wireframing, dynamic dashboards, SaaS agents, and image sorting.[1]
  • Based on Gemini 3 Pro architecture, optimized for low-latency, cost-sensitive enterprise use cases including transcription and data extraction.\[2]\[4]

Impact

Google's Gemini 3.1 Flash-Lite intensifies competition in the lightweight AI model segment by undercutting rivals on price and speed, with input costs at $0.25 per million tokens positioning it below OpenAI's GPT-4o Mini ($0.15 input but higher overall for comparable tiers) and Anthropic's Claude 3 Haiku, while delivering superior benchmarks like 1432 Elo on Arena.ai that match or exceed these models in reasoning and multimodality. This launch expands Google's tiered Gemini lineup—from Ultra to Flash-Lite—mirroring cloud providers' strategy to capture diverse developer needs, potentially accelerating adoption in high-volume enterprise applications like real-time agents and moderation where latency under 2.5x prior models matters. By embedding adjustable thinking levels natively, it lowers barriers for on-device-like inference at scale, aligning with trends in agentic workflows and cost-optimized R&D. Over the next 12-24 months, expect this to pressure funding toward efficient frontier models, narrowing gaps in API ecosystems and steering hyperscalers toward hybrid reasoning architectures amid GPU constraints.

Read more