Gemini 3.1 Flash-Lite: Google's $0.25/M Token Model Beats GPT-5 Mini

The Price War Has a New Leader

On March 3, 2026, Google released Gemini 3.1 Flash-Lite — the fastest and most cost-efficient model in the Gemini 3 family. At $0.25 per million input tokens and $1.50 per million output tokens, it's designed to make frontier AI capabilities accessible for high-volume, latency-sensitive workloads.

This pricing puts Flash-Lite at roughly 1/8th the cost of Gemini 3.1 Pro, making it the cheapest frontier-quality model available from any major provider.

Speed That Actually Matters

Flash-Lite isn't just cheap — it's fast. Key performance metrics:

• 45% faster overall answer generation compared to Gemini 2.5 Flash
• 2.5x shorter time to first token — meaning users see responses start almost instantly
• 299.6 tokens per second output speed, well above average for its price tier
• 1 million token context window with up to 64,000 output tokens

For applications like real-time translation, content moderation, and classification, these latency improvements translate directly to better user experiences and lower infrastructure costs.

Benchmark Surprise: Beating Bigger Models

Google ran 11 benchmark tests, and Flash-Lite topped 6 of 11 — beating both OpenAI's GPT-5 mini and Anthropic's Claude 4.5 Haiku on those tests. On the Artificial Analysis Intelligence Index, Flash-Lite scores 34, nearly double the median score (19) for models in its price tier.

The quality matches Gemini 2.5 Flash — a model that costs significantly more — while providing substantial improvements over previous Lite variants. This is the "quality floor rising" trend in action: today's budget models match last generation's premium offerings.

Four Thinking Levels

Flash-Lite supports configurable thinking levels — minimal, low, medium, and high — letting developers balance response quality against speed for each request. Need fast classification? Use minimal. Need nuanced analysis on a budget? Crank it to high.

This flexibility makes Flash-Lite viable for a surprisingly wide range of tasks, from simple API calls to moderate reasoning workloads that previously required more expensive models.

Where It Fits in Your Stack

Flash-Lite is available in preview via Google AI Studio (free tier) and Vertex AI (enterprise). Ideal use cases include:

• Content moderation at scale — process millions of items without budget anxiety
• Translation — near-instant multilingual conversion with frontier quality
• Classification and tagging — high-throughput document processing
• UI generation — create interface components with fast iteration cycles
• Simulation — rapid scenario generation for testing and planning

The Bigger Picture

Flash-Lite completes Google's Gemini 3.1 tiered strategy: Pro for maximum capability, Flash for balanced performance, and Flash-Lite for maximum throughput at minimum cost. This mirrors OpenAI's Instant/Thinking/Pro split and Anthropic's Haiku/Sonnet/Opus tiers — the industry has converged on offering models at every price-performance point.

TL;DR

Gemini 3.1 Flash-Lite costs $0.25/M input tokens, runs 45% faster than 2.5 Flash, and beats GPT-5 mini on 6/11 benchmarks. With 1M context and 4 thinking levels, it's the best value in frontier AI.

—

Compare model pricing and capabilities in our LLM pricing calculator and explore the full models comparison page.