TL;DR: Flash-Lite ($0.25/$1.50) is the cheap high-volume workhorse. 3.5 Flash ($1.50/$9) is the new default and tops 3.1 Pro on coding and agents. 3.1 Pro ($2-4/$12-18) is for the hardest reasoning. Newest is not priciest.
Google has three Gemini 3 models worth comparing in 2026. Gemini 3.1 Flash-Lite is the cheapest at $0.25 input and $1.50 output per million tokens. Gemini 3.5 Flash is the newest, at $1.50 and $9, and Google says it beats the older Pro flagship on coding and agent work. Gemini 3.1 Pro sits on top for the hardest reasoning and longest context, at $2 to $4 input and $12 to $18 output. All three read a 1M token context window. The surprise is that the newest model is not the most expensive.
Key takeaways
- Three Gemini 3 tiers: 3.1 Flash-Lite ($0.25/$1.50), 3.5 Flash ($1.50/$9), 3.1 Pro ($2-4/$12-18). All read 1M tokens.
- On a sample 100M-in, 50M-out month, the bill runs $100 on Flash-Lite, $600 on 3.5 Flash, and $800 on 3.1 Pro.
- 3.5 Flash is the new default. Google says it tops 3.1 Pro on coding and agent benchmarks while running about 4x faster.
- 3.1 Pro still wins the hardest reasoning: ARC-AGI-2 77.1%, GPQA Diamond 94.3%, SWE-Bench Verified 80.6%.
- Flash-Lite is built on Gemini 3 Pro, not on Flash. It is a Pro-grade model tuned for speed and high volume.
The three Gemini models at a glance
Here is the head-to-head, with the exact model name your engineers will call in the API. Use this as your reference. Everything below explains when each one is the right call. For the full lineup, including the 2.5 line and the media models, see our guide to every Gemini AI model compared.
| Model | Released | Context / max output | Price in / out (per 1M) | Best for |
|---|---|---|---|---|
| Gemini 3.1 Flash-Lite | Mar 3, 2026 | 1M / 64K | $0.25 / $1.50 | High-volume, low-latency tasks at scale |
| Gemini 3.5 Flash | May 19, 2026 | 1M / 64K | $1.50 / $9.00 | The new default for coding and agents |
| Gemini 3.1 Pro | Feb 19, 2026 | 1M / 64K | $2-4 / $12-18 | Hardest reasoning and long-context work |
Pro pricing has two tiers. You pay $2 input and $12 output for prompts up to 200K tokens, and $4 input and $18 output above that. Flash-Lite and 3.5 Flash hold one flat rate. Here is Google DeepMind’s own launch video for the Gemini 3 family.
Which Gemini model fits your next build?
Pick the job you have in mind to see the right model.
For translation, content moderation, classification, and routing at scale, Flash-Lite is the cheapest pick at $0.25/$1.50. It is built on Gemini 3 Pro, so quality holds up at high volume.
For daily coding and agent workflows, 3.5 Flash is the new default at $1.50/$9. Google says it tops 3.1 Pro on coding and agent benchmarks. Pair it with AI agent developers who ship on Gemini.
For deep research, algorithmic work, and the hardest reasoning, 3.1 Pro leads with ARC-AGI-2 77.1% and GPQA Diamond 94.3%. Keep it for the small slice of work that truly needs it.
The model is the easy part. The hard part is the team that builds and ships with it. We match you with pre-vetted engineers in about 24 hours. Tell us what you need.
The price gap, and why newest is not priciest
Per-million prices are abstract. A monthly bill is not. Take a product that reads 100 million input tokens and writes 50 million output tokens a month. The chart below shows the all-in model bill on each tier for that exact workload.
The same work costs $100 on Flash-Lite, $600 on 3.5 Flash, and $800 on 3.1 Pro. Output is the driver, since most apps read a short prompt and write a longer answer. The gap between the cheapest and the dearest is 8x on output price alone. That is why routing matters more than picking one model and using it for everything.
The order may surprise you. Gemini 3.5 Flash launched after 3.1 Pro, yet it costs less. Google priced the newest model as a Flash tier, not a Pro tier, even though it matches or beats Pro on many jobs. Two more levers shrink any of these bills. The Batch API runs non-urgent jobs at half price. Context caching stores a large fixed prompt once and reuses it cheaply. A team that uses both serves heavy workloads for far less than the sticker rate.
Gemini 3.1 Flash-Lite: the high-volume workhorse
Gemini 3.1 Flash-Lite is Google’s fastest and most cost-efficient Gemini 3 model. It launched on March 3, 2026, at $0.25 input and $1.50 output per million tokens. It reads up to 1M tokens of context and writes up to 64K tokens. It is built for high-volume developer work where speed and price matter more than peak reasoning.
Two facts make Flash-Lite stand out. First, it is fast. Google reports a 2.5x faster time to first answer token and a 45% jump in output speed versus the older 2.5 Flash. Second, it is built on Gemini 3 Pro, not on a smaller Flash base. So you get Pro-family quality tuned for throughput. On GPQA Diamond it scores 86.9%, strong for a model this cheap.
Google points Flash-Lite at high-volume translation, content moderation, classification, and instruction following. It is also good for generating user interfaces, dashboards, and simulations. A common pattern is to use it as the router or orchestrator in an agent system. It reads each request, decides what to do, and only hands the hard cases up to a bigger model. That keeps most of your traffic on the cheapest tier.
Gemini 3.5 Flash: the new default for coding and agents
Gemini 3.5 Flash is the first model in the 3.5 family. Google launched it at I/O 2026 on May 19 and made it the default in the Gemini app and AI Mode worldwide. It costs $1.50 input and $9 output per million tokens, reads 1M tokens, and writes up to 64K. Google describes it as its strongest model yet for agents and coding.
The headline is speed. Google says 3.5 Flash runs about 4x faster than other frontier models in output tokens per second. On its own benchmarks it scores 76.2% on Terminal-Bench 2.1, 55.1% on SWE-Bench Pro, and 83.6% on the MCP Atlas tool-use test. Google states it surpasses 3.1 Pro on coding, agentic, and multimodal benchmarks. So the newest Flash beats the older flagship Pro on the jobs most teams care about.
This is why 3.5 Flash is the new default. It is fast enough for live coding tools and chat, smart enough for multi-step agents, and priced 25% below Pro on output. Google calls out one strength in particular. It is built for deploying many agents at once and for long-running tasks. A fintech team we work with moved its coding agents from Pro to 3.5 Flash and cut latency in half with no drop in pass rate.
Gemini 3.1 Pro: for the hardest reasoning and longest context
Gemini 3.1 Pro launched on February 19, 2026, and it is still the top model for raw reasoning. It reads 1M tokens, writes up to 64K, and costs $2 to $4 input and $12 to $18 output depending on prompt size. If a Flash model keeps failing a task, Pro is where you go next.
The benchmarks back this up. Gemini 3.1 Pro scores 77.1% on ARC-AGI-2, a hard abstract-reasoning test, which is more than double what Gemini 3 Pro scored three months earlier. It hits 94.3% on GPQA Diamond for graduate-level science and 80.6% on SWE-Bench Verified for agentic coding. On competitive coding it reaches 2887 Elo on LiveCodeBench Pro.
Pro is the pick for deep research, complex multi-step reasoning, long-context synthesis, and algorithmic development. The catch is cost and speed. Output is 25% to 100% dearer than 3.5 Flash, and it runs slower. For most teams, Pro should be a small slice of total traffic, reserved for the hardest jobs, not the daily driver. Send the routine work to Flash or Flash-Lite and pay the Pro rate only when you must.
Benchmarks, side by side
One warning before the table. Google does not run every model on the exact same test, and benchmark versions change. So read each score next to its own test name, not as a clean head-to-head. The pattern still holds. Pro leads pure reasoning, 3.5 Flash leads coding and agents, and Flash-Lite holds its own for the price.
| Benchmark | 3.1 Flash-Lite | 3.5 Flash | 3.1 Pro |
|---|---|---|---|
| GPQA Diamond (science) | 86.9% | — | 94.3% |
| ARC-AGI-2 (reasoning) | — | — | 77.1% |
| SWE-Bench (agentic coding) | — | 55.1% (Pro, public) | 80.6% (Verified) |
| Terminal-Bench (terminal coding) | — | 76.2% (v2.1) | 68.5% (v2.0) |
| MCP Atlas (tool use) | — | 83.6% | — |
| MMMU-Pro (multimodal) | 76.8% | 83.6% | — |
Notice Terminal-Bench. Gemini 3.5 Flash scores 76.2% on version 2.1, while 3.1 Pro scores 68.5% on version 2.0. The versions differ, so this is not a perfect match, but it shows how far the newest Flash has come on agentic coding. This is the core story of 2026. A cheaper, faster model now does the coding and agent work that used to need the flagship.
How to choose, fast
Two questions settle most decisions. How hard is the task, and how high is the volume. Plot those and the right model falls out. High-volume simple work goes to Flash-Lite. Normal coding and agents go to 3.5 Flash. Only the hardest reasoning justifies Pro.
One more lever sits inside each model. All three Gemini 3 models support thinking levels, so you can tell the model how hard to reason on a given call. A low thinking level is fast and cheap for simple jobs. A high level digs deeper on hard ones. This means you do not always have to jump to a bigger model when a task gets harder. Often you can raise the thinking level on 3.5 Flash first, and only move to Pro if that still falls short.
The table below maps common jobs to a model so you can decide in seconds. Notice how much sits on 3.5 Flash. That is by design. The default tier is the right answer far more often than teams assume, and it is cheaper than the Pro habit most teams started with.
| Job | Recommended model | Why |
|---|---|---|
| Translation, moderation, classification | 3.1 Flash-Lite | Cheapest, fast, high volume |
| Routing or orchestrating an agent system | 3.1 Flash-Lite | Pro-grade quality at the lowest price |
| Daily coding and code review | 3.5 Flash | Beats Pro on coding, 25% cheaper |
| Multi-agent and long-running tasks | 3.5 Flash | Built for agents, about 4x faster |
| Deep research, long-context synthesis | 3.1 Pro | Top reasoning and long-context scores |
| Hardest reasoning, algorithmic work | 3.1 Pro | ARC-AGI-2 and GPQA leader |
What this means for hiring
Knowing the three models is step one. Getting value from them is the harder step. The teams that win are not the ones running Pro on everything. They route work to the right tier, cache context, and ship a clean product around the model. That is engineering, not a model choice.
Strong AI engineers are scarce and costly in the United States. A senior AI engineer there runs $12,000 to $18,000 per month all in. The same skill from a pre-vetted AI and machine learning engineer in Vietnam or the Philippines often costs $3,000 to $6,000 per month. The work is the same. The bill is not. Our Vietnam versus US developer cost breakdown shows the full math.
You can size any role on our developer rate card. The model lineup will keep changing. Google shipped three Gemini 3 models in four months this year, and a 3.5 Pro is likely next. The need for engineers who route, cache, and ship well will not change. That is the skill that turns a cheap model bill into a working product.
Build your Gemini stack with the right team
Gemini gives you three strong models and clear prices. The choice between them is easy once you match the model to the job. Flash-Lite for volume, 3.5 Flash for coding and agents, 3.1 Pro for the hardest reasoning. The team that builds with them is what ships value. We match you with pre-vetted senior engineers across Asia in about 24 hours, with no upfront cost and payroll handled.
