Gemini 3.5 Flash vs Gemini 3.1 Flash-Lite: Cost, Speed, Coding, and High-Volume Tasks
Gemini 3.5 Flash and Gemini 3.1 Flash-Lite sit six times apart on price, and the gap in what they're built for is just as wide — one is Google's coding and agent workhorse, the other is built to run the same simple task millions of times as cheaply as possible.
........
Gemini 3.5 Flash lists at $1.50 per million input tokens and $9.00 output; Gemini 3.1 Flash-Lite lists at $0.25 input and $1.50 output — roughly 6x cheaper on both ends.
Both share Google's standard 1,048,576 token context window, with Flash-Lite capped at 65,536 output tokens per response.
Google says Gemini 3.5 Flash runs about 4x faster than other frontier models in output tokens per second, and surpasses Gemini 3.1 Pro on coding, agentic, and multimodal benchmarks.
Gemini 3.1 Flash-Lite scores 34.7 on a coding index and 25.6 on an aggregate intelligence index — noticeably behind Flash, as its tier positioning implies.
Flash-Lite has already been superseded once, by Gemini 3.5 Flash-Lite, which posted 23-point gains on agentic terminal tasks over the 3.1 generation.
··········
WHERE EACH MODEL SITS IN THE LINEUP.
Google runs three tiers per Gemini generation — Pro, Flash, and Flash-Lite — and this comparison sits two tiers apart rather than one.
Gemini 3.5 Flash is the workhorse tier: the model Google positions for daily coding, agent workflows, and production tasks that need real reasoning under a moderate budget.
Gemini 3.1 Flash-Lite sits at the bottom of the ladder, optimized specifically for low-latency, high-volume workloads — the kind of task that runs constantly and needs to stay cheap more than it needs to be clever.
That is not a small gap. Flash-Lite is the tier built to be outgrown; Flash is the tier built to be the default. Comparing them head-to-head only makes sense once that difference in intent is on the table.
Google's own material frames the decision around two questions: how hard is the task, and how high is the volume. High-volume, simple, easily-verified work belongs on Flash-Lite. Everyday coding and multi-step agent work belongs on Flash.
··········
PRICING.
The price difference is not incremental — Flash-Lite runs at roughly a sixth of Flash's rate on both input and output.
Gemini 3.5 Flash is priced at $1.50 per million input tokens and $9.00 per million output tokens.
Gemini 3.1 Flash-Lite is priced at $0.25 per million input tokens and $1.50 per million output tokens — reported consistently across multiple pricing trackers, though a couple of secondary sources show the input rate drifting between $0.125 and $0.45 depending on the provider and date checked, so it's worth confirming the live rate before budgeting at scale.
Both tiers share Google's flat-pricing design on the Flash line: unlike the Pro tier, which doubles its rate for prompts over 200,000 tokens, Flash and Flash-Lite pricing stays flat regardless of prompt length.
Cache reads bring the cost down further on both: Flash-Lite's cache read rate is roughly a tenth of its standard input price, in line with Google's general cache-read discount across the Gemini 3 family.
··········
Pricing per million tokens
Gemini 3.5 Flash | Gemini 3.1 Flash-Lite | |
|---|---|---|
Input | $1.50 | $0.25 |
Output | $9.00 | $1.50 |
Context window | 1,048,576 tokens | 1,048,576 tokens |
Max output | Not separately published | 65,536 tokens |
Long-context surcharge | None (flat rate) | None (flat rate) |
··········
WHAT $9 VS $1.50 ACTUALLY COSTS AT SCALE.
Small per-token gaps compound fast once volume enters the picture, and this gap is not small.
A monthly workload of 20 million input tokens and 2 million output tokens is a reasonable proxy for a moderate production pipeline.
On Flash-Lite pricing, that workload runs to roughly $8 for input and $3 for output — call it around $11 total before any caching discount.
On Flash pricing, the same volume runs to $30 for input and $18 for output — around $48 total, more than four times as much for identical token counts.
That gap is exactly why the tier choice matters more than prompt-level optimization for high-volume applications: no amount of trimming a prompt closes a 6x rate difference.
··········
BENCHMARKS.
The two models were not tested on an identical suite, so any comparison has to be read benchmark-by-benchmark rather than as a single score.
Google reports Gemini 3.5 Flash scoring 76.2% on Terminal-Bench 2.1, 55.1% on SWE-Bench Pro, and 83.6% on the MCP Atlas tool-use test, and states that it surpasses Gemini 3.1 Pro — the previous generation's flagship — on coding, agentic, and multimodal benchmarks.
Gemini 3.1 Flash-Lite's published scores sit in a different set: 34.7 on a coding index, 82.2% on GPQA Diamond, and an aggregate intelligence score of 25.6, placing it in roughly the 67th percentile among models tracked by one independent aggregator.
One data point that does line up directly: Gemini 3.1 Flash-Lite recorded 31.0% on the same Terminal-Bench 2.1 suite where Flash scores 76.2% — a 45-point gap on the one benchmark both were run against in the sources reviewed here.
That gap is consistent with the tier gap in pricing: Flash-Lite is not trying to compete with Flash on reasoning-heavy or agentic tasks, and its own successor model was measured against exactly that shortfall.
··········
Where benchmark data overlaps
Benchmark | Gemini 3.5 Flash | Gemini 3.1 Flash-Lite |
|---|---|---|
Terminal-Bench 2.1 | 76.2% | 31.0% |
SWE-Bench Pro | 55.1% | Not directly published |
Coding index (aggregator) | Not directly published | 34.7 |
GPQA Diamond | Not directly published | 82.2% |
··········
SPEED.
Both models are built for low latency, but Google's own comparison puts Flash meaningfully ahead in raw throughput.
Google states that Gemini 3.5 Flash runs at roughly four times the output speed, in tokens per second, of other frontier models — a speed claim aimed at justifying its use in live coding tools and interactive chat.
Gemini 3.1 Flash-Lite is designed around the same low-latency priority but for a different reason: it's meant to be the cheapest, fastest option for tasks run constantly rather than the fastest option for a single complex exchange.
In practice, the meaningful speed comparison for most teams isn't Flash against Flash-Lite directly, but each of them against the task they're actually assigned: Flash-Lite for simple, repeated, verifiable calls; Flash for anything that needs a real reasoning step before responding.
··········
WHAT EACH ONE IS ACTUALLY FOR.
The clearest way to choose between them is by what happens when the model gets something wrong.
Flash-Lite is the right default when errors are cheap and easy to catch: classification, simple extraction, translation, routing, and any task run at high volume where a wrong answer gets flagged and retried automatically.
Flash is the right default when a task needs multiple steps, tool calls, or actual reasoning before producing something usable — daily coding work, agent workflows that chain several actions together, and anything where Google's own benchmarks show a real capability gap.
Document processing, search, and other high-throughput developer workflows are the use cases Google explicitly built Flash-Lite around, and they're a good filter: if the task looks like something you'd run a million times a day without reading each result individually, that's a Flash-Lite task.
··········
A GENERATION GAP WORTH KNOWING ABOUT.
Both of these models have already been followed by newer versions, which matters for anyone planning a deployment now rather than reading about a snapshot.
Gemini 3.1 Flash-Lite has been succeeded by Gemini 3.5 Flash-Lite, released July 21, 2026 at the same $0.30/$2.50 price point roughly, with Google reporting 16 to 23 percentage point jumps on coding, computer-use, and terminal benchmarks over the 3.1 generation — Terminal-Bench nearly doubling, from 31.0% to 54.0%.
Gemini 3.5 Flash itself has been followed by Gemini 3.6 Flash, which lowered the output price to $7.50 per million tokens while improving benchmark scores further.
Neither Gemini 3.5 Flash nor Gemini 3.1 Flash-Lite is the current top of its respective sub-tier as of this writing — both are one generation behind their direct successors, which are worth checking before committing to either for a new project.
··········
HOW TO CHOOSE.
The practical rule holds regardless of which generation is current: match the model to the cost of being wrong, not to the lowest price on the page.
Start on Flash-Lite for any workload that is high-volume and easy to verify — the price difference at scale is too large to ignore, and the benchmark gap matters less when errors are cheap to catch.
Move to Flash for anything involving multiple steps, tool use, or a task where a single bad output is expensive to unwind — a coding agent, a multi-turn workflow, or a task with no easy way to check the answer automatically.
The one number that actually settles the decision is cost per accepted result on your own workload, not the sticker price or a single benchmark row: a cheap model that needs three retries can end up costing more than the expensive one that got it right the first time.
Given that both models already have newer successors available at similar or better pricing, it's worth testing the current-generation equivalents — Gemini 3.5 Flash-Lite and Gemini 3.6 Flash — alongside these two before finalizing a choice for new work.
··········
·····
FOLLOW US FOR MORE.
·····
·····
DATA STUDIOS
·····
