Gemini 3.8 Flash

Why this oneWhat it is actually for

Reach for gemini-3.8-flash when you want Google's Flash economics on work that has outgrown a Flash model — the vendor positions it as "our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows".

It is also the row that makes Google's board honest. The two Gemini entries this site carried before it — gemini-3.1-pro and gemini-3-flash — are preview channels on the vendor's own page, not stable ids. This is the stable one.

What it isArchitecture, lineage, training

Gemini 3.8 Flash is a closed-weight model in Google's Flash line, marked on the vendor's pricing page as a current stable release rather than a preview. It sits above 3.7 Flash and 3.6 Flash, which carry the same $0.75/$3.75 rate and the same 2027 increase, and above 3.5 Flash at $1.50/$9.00.

⚠︎ The defining fact about this row is its expiry date. Google publishes the rate as good through 31 December 2026 and $1.50/$7.50 from 1 January 2027 — a doubling that is already announced. This board carries that date in the row's expires field, which is the alarm the daily currency watch reads each morning.

At a glanceSee it

The only model on this board whose price rise is already published — which is why its row carries an expiry date rather than a null.

CapacityContext, output and what fits

The parametersEvery knob, and what moving it does

Google's per-model parameter documentation was not reachable from the pages read for this entry, so no sampling, thinking or tool parameters are asserted here.

SamplingShaping the output distribution

Not documented on the pricing page or the models guide read for this entry. This page does not restate another Gemini model's controls as though they applied here — check Google's per-model reference before relying on a specific range.

ReasoningThinking, effort and budgets

Google positions this model for "long-horizon software engineering, autonomous agents, and complex enterprise workflows", which implies sustained multi-step work, but the pages read here document no reasoning-effort control by name and none is claimed.

ToolsFunction calling and server tools

Not documented on the pages read for this entry. The Flash line generally carries function calling, structured outputs and context caching, but that is an inference from the family rather than a fact about this id, so it is recorded as unknown rather than asserted.

CostPrice, caching, batching, what drives the bill

List price from Google's pricing page, per million tokens:

The board quotes the current rate and carries 31 December 2026 as the row's expiry, so the increase is an alarm the daily watch raises rather than a surprise on a bill. Gemini 3.7 Flash and 3.6 Flash carry the same pair of rates and the same date.

Where it runsSurfaces and availability

StrengthsWhat it is good at

• A stable id, where the two Gemini rows this board carried before it are preview channels.
• Flash-tier pricing on work Google positions for agents and long-horizon engineering.
• Half the price of Gemini 3.5 Flash on input and well under half on output.
• The price rise is published in advance rather than discovered on an invoice.

LimitsWhere it falls down

• The price doubles on 1 January 2027, and that is already announced.
• Context window, max output, knowledge cutoff, tool support and sampling controls were not reachable on the vendor pages read here, so this page asserts none of them.
• Newer than the rows around it, with correspondingly less independent evidence.

Against its neighboursHow it compares

Against gemini-3-flash at $0.50/$3.00: cheaper on paper, but the vendor page carries it only as gemini-3-flash-preview — a preview channel, not a stable id. That is the trade, and it is about channel stability rather than price.

Against claude-haiku-4-5 at $1/$5 and openai/gpt-5-6-luna at $0.20/$1.20: this row sits between them, above Luna and below Haiku, until 1 January 2027 moves it above both.

Getting startedThe smallest call that works

Pin the stable id rather than a -preview channel, and put 1 January 2027 in your own calendar as well as trusting this board's expiry field. If your workload is priced on this row, model the post-2027 rate now: a doubling that is already published is a budgeting fact, not a risk.

SourcesWhere every claim above came from

• Google AI pricing — ai.google.dev/pricing (read 18 Sep 2026)
• Gemini API models guide — ai.google.dev/gemini-api/docs/models (read 18 Sep 2026)

RelatedGo deeper