Skip to content

Gemini 3: Pro, Flash, Deep Think, Pricing and Access

A woman looking at a laptop showing a bright blue and purple starburst in a sunny home office


A woman looking at a laptop showing a bright blue and purple starburst in a sunny home office

Gemini 3 is Google’s third-generation family of multimodal AI models, led by Gemini 3.1 Pro, and it now sits one generation below the Gemini 4 models that top Google’s lineup.

Google DeepMind built Gemini 3 to reason across text, images, video, audio, and code in one model. At launch, Google said Gemini 3 Pro topped the LMArena leaderboard at 1501 Elo, the first model past 1500. Google’s enterprise announcement lists multimodal understanding, agentic coding, and tool use among the strengths it shipped with.

What Gemini 3 is, and where it sits now

Gemini 3 is a family, not one model: Pro for the hardest work, Flash for everyday speed, and Flash-Lite for high-volume jobs at the lowest cost. The current models are Gemini 3.1 Pro, Gemini 3.8 Flash, and Gemini 3.5 Flash-Lite. The original 3 Pro preview was replaced by 3.1 Pro at the same price.

Google’s pricing page, checked at writing time, leads with Gemini 4 Carbon, called its most capable reasoning model, then Gemini 4 Argon, the efficient Gemini 4 frontier model. Gemini 3.1 Pro sits further down as the third-generation Pro model. In the Gemini app, the picker lists Flash-Lite, Flash, Pro, and Argon without version numbers. The Gemini vs ChatGPT comparison compares the two lineups.


Three stacked blocks in blue, purple, and teal with a glowing gold star above the top block

Pro vs Flash vs Flash-Lite: which model to pick

All three tiers take up to 1 million tokens of input and accept text, images, audio, video, and files. They differ in reasoning depth, speed, and price. HaiAi123 lists the family tiers with their model IDs and per-token prices.

Pro is the deep thinker. Its API model ID is gemini-3.1-pro-preview, and it is still in preview, so Google can still change it. It suits hard reasoning and complex coding. It costs $2 per million input tokens and $12 per million output tokens for prompts up to 200,000 tokens, rising to $4 and $18 past that line, with no free API tier.

Flash is the workhorse. The original Flash preview cost $0.50 input and $3 output per million tokens. The current model, Gemini 3.8 Flash, costs $0.75 and $3.75 through December 31, then $1.50 and $7.50 from January 1, with a free tier. Start here for most work.

Flash-Lite is the cheapest tier, for simple high-volume calls: sorting, extraction, translation, and routing inside an agent. Gemini 3.1 Flash-Lite costs $0.25 and $1.50, and the current 3.5 Flash-Lite costs $0.30 and $2.50, with a free tier. Other labs split their models the same way, as the DeepSeek V4 lineup and the ChatGPT model lineup show.


A laptop showing three blank plan cards on a wooden desk beside a notebook and a cup of coffee

Deep Think: slow reasoning for hard problems

Deep Think is a reasoning mode on Pro, not a separate model. Switched on, the model spends far more compute before it answers, working through several lines of thought at once and checking its own logic first. That is why an answer can take minutes instead of seconds. The Rundown describes Deep Think as Google’s most compute-intensive reasoning mode, built on Gemini 3.1 Pro.

Use it where a wrong answer costs more than a wait: hard science, math, logic, and engineering design. Google reported 91.9% for Gemini 3 Pro on GPQA Diamond, a test of PhD-level science questions.

Access has moved between Google’s paid plans, so check the plan page before you subscribe for it. The trade stays the same: more thinking, slower answers, better results on hard problems.


Glowing blue and purple branching paths spreading across a dark background

Multimodal input and the 1 million token window

Gemini 3 reads text, images, video, audio, code, and documents together in one prompt. Hand it a video, a photo of a blurry receipt, or hours of meeting audio, and ask questions about what is in them.

The context window holds 1 million tokens, roughly 750,000 words: several books, a whole codebase, or hours of video in one request. The catch on Pro is price: prompts past 200,000 tokens bill at the higher rate, so test long prompts on Flash first.

The knowledge cutoff is January 2025, so anything newer needs grounding with Google Search. Google includes 5,000 grounding requests a month, shared across Gemini 3 and newer models, then charges $14 per 1,000.


A laptop on a desk surrounded by a photo, a video frame, a sound wave, and a document page connected by glowing lines

Agentic coding and front-end work

Gemini 3 was built to act, not just answer. It plans multi-step tasks, calls tools, edits files, and checks its own work. Google ships it in Gemini CLI and Antigravity, and it runs inside Cursor, GitHub Copilot, JetBrains tools, and Replit.

Its most visible trick is front-end work from one prompt: describe an interface and get a working prototype. It also helps move old code to a new framework and write tests.

Google reported 76.2% for Pro and 78% for Flash on SWE-bench Verified, a test of real software fixes. Treat those as vendor numbers and test on your own code.


A developer at two monitors showing colorful abstract waves in an evening office

Pricing and where to use it

API pricing is per million tokens. Pro costs $2 input and $12 output up to 200,000 tokens, then $4 and $18. Gemini 3.8 Flash costs $0.75 and $3.75 through December 31, then $1.50 and $7.50. Gemini 3.5 Flash-Lite costs $0.30 and $2.50. For scale, Gemini 4 Argon lists at $2 and $10 through November 30, then doubles. Google lists the current rates, free tier status, and dated price steps on its Gemini API pricing page.

In the Gemini app, free accounts now get automatic routing after an October reshuffle, which sends most prompts to Flash-Lite. Google AI Pro costs $19.99 a month and opens Flash, Pro, and Argon.

You can use Gemini 3 in the Gemini app, in AI Mode in Google Search, in AI Studio, and in Vertex AI and Gemini Enterprise. Try it first in AI Studio, and read the Gemini Enterprise guide for team use. The best AI tools compared puts Gemini 3 beside the other models worth testing.


A phone, a laptop, and a tablet showing blank chat bubbles with blue and purple glows on a wooden desk

Frequently asked questions about Gemini 3

What is Gemini 3?

Gemini 3 is Google’s third-generation family of multimodal AI models. It has Pro, Flash, and Flash-Lite tiers and takes up to 1 million tokens of context.

Is Gemini 3 still Google’s best model?

No. Google’s pricing page now leads with Gemini 4 Carbon and Gemini 4 Argon. Gemini 3.1 Pro is the third-generation Pro model, one generation behind.

Which Gemini 3 model should I use, Pro, Flash, or Flash-Lite?

Use Flash for most work. Use Flash-Lite for high-volume simple jobs like sorting and extraction. Use Pro for hard reasoning and complex coding.

What is Deep Think?

Deep Think is a reasoning mode on Gemini 3 Pro that spends more compute exploring several lines of thought before answering. It suits hard math, science, and logic problems.

How much does Gemini 3 cost?

On the API, Gemini 3.1 Pro costs $2 per million input tokens and $12 per million output tokens up to 200,000 tokens. Flash and Flash-Lite cost less and have free tiers.

Where can I use Gemini 3?

In the Gemini app, in AI Mode in Google Search, in AI Studio, in Vertex AI and Gemini Enterprise, and through coding tools such as Gemini CLI and Antigravity.

Leave a Comment