On March 3, 2026, Google announced Gemini 3.1 Flash Lite in preview, available through the Gemini API in AI Studio and on Vertex AI. The positioning is blunt: the fastest and most cost-efficient model in the Gemini 3 series, built for high-volume developer workloads at scale. The launch also breaks from convention — the Gemini 3 generation originally skipped Flash-Lite entirely, and this cycle Google jumped past a standard Flash to ship the Lite variant first, roughly two weeks after Gemini 3.1 Pro arrived on February 19. The sequencing itself says something about where Google sees demand: cheaper inference tokens, sooner.
VentureBeat pegs its cost at about one-eighth of Pro. For products that bill by the token, this is not a minor revision — it adds a new cell to the model-selection matrix, and one that many high-volume pipelines will find hard to ignore.
Pricing and Speed
The price is $0.25 per million input tokens and $1.50 per million output tokens, the lowest in the Gemini 3 family. Against the previous generation, Gemini 2.5 Flash-Lite sat at $0.10 and $0.40 — the new model costs several times more than its predecessor, and that gap is the premium charged for third-generation capability. Anyone budgeting a migration should model it explicitly rather than assume “Lite” means cheap in absolute terms.
On speed, The New Stack cites up to 363 tokens per second, the fastest of any Gemini 3 model — three tokens per second slower than the 2.5 Flash-Lite predecessor, yet two to five times faster than rivals in the same tier. Google’s own numbers, attributed to Artificial Analysis, claim a 2.5x faster time to first answer token and 45% higher output speed versus 2.5 Flash, at similar or better quality. For interactive products where time-to-first-token drives perceived latency, that combination — cheaper than Pro, faster than the field — is the actual product.
Benchmarks and the Competition
On the Arena.ai leaderboard, Flash Lite posts an Elo of 1432. It scores 86.9% on GPQA Diamond and 76.8% on MMMU Pro, beating similar-tier models and even some larger prior-generation Gemini models on reasoning and multimodal understanding. The New Stack’s comparison rundown: it generally beats Gemini 2.5 Flash while costing less; direct competitors GPT-5 mini and Claude 4.5 Haiku come out behind; Grok 4.1 Fast is cheaper but markedly slower.
Notably, Google published no agent benchmarks — given how the rest of the market now positions small models as agent workhorses, the omission is itself a signal about where Google thinks this model does and does not belong. Also worth knowing: the consumer Gemini app will not carry this model. It is purely an API product, which keeps the consumer experience and the developer cost story on separate tracks.
Adjustable Thinking Levels
Flash Lite ships with adjustable thinking levels: developers dial how deeply the model reasons, per task. In high-frequency workloads this is the key cost lever — drop the reasoning level, emit fewer tokens, and the bill falls accordingly. It converts reasoning depth from a fixed property of the model into a per-request engineering decision, the same lever OpenAI offers with reasoning-effort parameters.
Google’s demos include filling an e-commerce wireframe with hundreds of products, generating a real-time weather dashboard from live and historical data, running a multi-step SaaS business agent, and rapidly analyzing and sorting large image sets. Early adopters include Latitude, Cartwheel, and Whering, with feedback centering on efficiency and instruction adherence — the two properties that matter most when a model sits at the bottom of a high-volume pipeline.
How to Choose
A practical split. High-volume data processing, translation, and content moderation, plus UI and dashboard generation and complex instruction following — that is Flash Lite’s home turf. For orchestration layers coordinating fleets of agents, Google itself offers no benchmark data; Pro or a purpose-built model is the safer pick until someone publishes credible agent numbers. If you currently run on 2.5 Flash-Lite, do the price math before migrating: cost rises several-fold, and what you buy is third-generation multimodal understanding and reasoning — worth it or not depending on how sensitive your traffic mix is to speed and quality. Specs may still shift during preview, and no general-availability date has been set, so pin versions in production and re-verify when the GA build lands.
Sources
- Gemini 3.1 Flash Lite — Google Blog
- Google Gemini 3.1 Flash Lite — The New Stack
- Google releases Gemini 3.1 Flash Lite at 1/8th the cost of Pro — VentureBeat
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.