Google's Gemini 3.1 Flash-Lite Is the Cheapest Capable Multimodal Model Yet

Google shipped Gemini 3.1 Flash-Lite on March 3rd in preview across the Gemini API, Google AI Studio, and Vertex AI. The pitch is straightforward: fastest time-to-first-token in Google’s lineup, cheapest price point, and benchmark numbers that beat both GPT-5 mini and Claude 4.5 Haiku on six of eleven evaluations.

At $0.25 per million input tokens, it’s worth taking seriously.

The Numbers

• $0.25/M input tokens — cheapest in Google’s current lineup
• 2.5x faster time-to-first-token vs Gemini 2.5 Flash
• 45% faster output throughput vs Gemini 2.5 Flash
• 1M token context window
• Up to 64K output tokens per call
• Multimodal — text, images, video, audio natively

The benchmark wins over OpenAI and Anthropic’s efficiency-tier models matter because this isn’t a toy model positioned below the serious ones. It’s competing directly at the tier where most production workloads actually run.

Who This Is Actually For

Flash-Lite is designed for high-volume production workloads — the use cases where you’re making millions of API calls and the per-token cost compounds fast. Specifically:

Chatbots and real-time interfaces. The 2.5x improvement in time-to-first-token is the difference between an interface that feels responsive and one that feels like it’s thinking. For user-facing products, this matters as much as raw capability.

Document processing pipelines. A 1M context window with 64K output capacity covers virtually any real-world document pipeline — legal review, financial analysis, technical documentation summarization — without chunking gymnastics.

Classification and routing. High-volume classification tasks (content moderation, intent routing, entity extraction) are where cost-per-call pricing hits hardest. Flash-Lite’s price point makes tasks that were previously marginal now clearly viable.

RAG at scale. The combination of context length and low cost makes it a natural fit for retrieval-augmented generation setups where you’re stuffing large retrieved chunks into every call.

How It Compares

The efficiency tier is now genuinely competitive across all three major providers. GPT-5 mini, Claude 4.5 Haiku, and Gemini 3.1 Flash-Lite are all capable enough for real production workloads and cheap enough to run at scale.

The differentiation is:

• Flash-Lite — fastest output, cheapest price, Google ecosystem integration
• GPT-5 mini — best OpenAI API ecosystem fit, strong on code tasks
• Claude 4.5 Haiku — strongest on instruction-following, best for agents that need precise output formatting

For most teams, the right answer is benchmark your specific workload against all three. The cost differences are real enough that the winner depends heavily on your actual task distribution.

Access

Gemini 3.1 Flash-Lite is available now in preview via the Gemini API at aistudio.google.com, Google AI Studio, and Vertex AI. Pricing is live — no waitlist.

---

Documentation: ai.google.dev — Gemini 3.1 Flash-Lite model card