Gemini 3.8 Flash — released September 2, 2026 — is Google’s newest Flash model as of September 4, 2026, and the $0.75 per million input tokens / $3.75 per million output tokens quoted everywhere is not its standard price. It is an introductory price that expires on December 31, 2026. Starting January 1, 2027, Google’s rate card moves to $1.50 input / $7.50 output per 1M tokens — a clean doubling on both dimensions. This page explains that mechanism, shows what the doubling does to a real bill, and covers the question most write-ups skip: how to call Gemini 3.8 Flash through an API today, and what switching looks like when the window closes. All figures were captured September 4, 2026; prices change without notice, so re-verify before committing spend.
One scope note first. If your search was really about generating images with a Gemini model, you want a different article: Gemini 2.5 Flash image generation runs on Modellix’s asynchronous media API, not on the text gateway this page covers. The two live on different hosts and are not interchangeable — that split is spelled out in our Gemini 2.5 Flash image API guide. We run Modellix, an API gateway listing Gemini 3.8 Flash beside 27 other language models, and have a commercial interest when Modellix appears below — we say so rather than hide it. Google and Modellix prices were both captured September 4, 2026.
Gemini 3.8 Flash pricing: $0.75 today, $1.50 on January 1, 2027
Google’s announcement (September 2, 2026) introduces Gemini 3.8 Flash “at $0.75 per million input tokens and $3.75 per million output tokens,” and is unusually specific about what happens next: “Introductory price expires on December 31, 2026. Starting January 1, 2027, $1.50/1M input tokens and $7.50/1M output tokens will apply.” Google does not disclose the context window on that page, and it gives no numeric performance deltas versus 3.7 Flash — the improvement language (“significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning”) stays qualitative. Both gaps matter for budgeting.
| Rate period | Input / 1M tokens | Output / 1M tokens | Source note |
|---|---|---|---|
| Introductory — through December 31, 2026 | $0.75 | $3.75 | Google announcement, Sept 2, 2026 |
| Standard — from January 1, 2027 | $1.50 | $7.50 | Same announcement: “Introductory price expires on December 31, 2026” |
The row on Modellix’s LLM price page mirrors Google’s numbers and is easy to misread, so here is the explanation. Each cell shows two prices with the first struck through — Input $1.50 → $0.75, Output $7.50 → $3.75 — and the row’s Discount column reads “50% OFF.” The struck-through $1.50/$7.50 is Google’s list price effective January 1, 2027; the $0.75/$3.75 is Google’s own introductory price. In other words, that “50% OFF” describes Google’s price schedule. It is not a Modellix discount. Modellix is a distribution layer, not the model maker: it neither adds to nor discounts Google’s numbers on this row, and when the introductory period ends the row’s price moves with it.
| Model (Modellix row, Sept 4, 2026) | Input (list → price) | Output (list → price) | Cached read (list → price) | Context window |
|---|---|---|---|---|
google/gemini-3.8-flash |
$1.50 → $0.75 | $7.50 → $3.75 | $0.15 → $0.075 | 1M tokens |
google/gemini-3.7-flash |
$1.50 → $0.75 | $7.50 → $3.75 | $0.15 → $0.075 | 1.05M tokens |
google/gemini-3.6-flash |
$1.50 → $1.35 | $7.50 → $6.75 | $0.15 → $0.135 | 1.05M tokens |
Source: Modellix LLM price page, captured September 4, 2026. “List” is the official list price shown struck through; “price” is what Modellix bills. All rates USD per 1M tokens.
Three details in that table deserve close reading:
- The context window did not grow. The Modellix table lists Gemini 3.8 Flash at a 1M-token window — smaller than 3.6 and 3.7 Flash, which list 1.05M. Google’s announcement discloses no window for 3.8 Flash, so the Modellix table value (1M) is the only number we will quote. If window size alone decides between them, the table says 3.7 Flash is larger.
- The Cached Write column lists $0, and that is not a feature claim. The table lists cached writes at $0, but the documentation does not state whether cache writes are supported on Gemini models — “$0” may mean “not billed” or “not applicable,” and we will not guess.
- Multimodal input is native to the model, not to the gateway. Gemini 3.8 Flash natively accepts text, image, video, and audio input (the Modellix table marks it text/image/video/audio → text). But the Modellix LLM gateway is a text gateway — the API reference instructs text-only prompts and no attached images — so plan for text-in/text-out through this endpoint.
Separately, Google also announced Gemini 3.8 Flash Cyber, a cybersecurity variant for trusted defenders only (Fairwind Program); it is not in Modellix’s catalog and out of scope here.
What the price doubling does to a real bill
Gemini 3.8 Flash is flat-rate — one input price and one output price per million tokens, no per-tier surcharge inside the context window — so the arithmetic is simple, and Modellix’s on-page cost calculator reproduces it.
Take a steady workload: 20 million input tokens and 2 million output tokens per month, the shape of a batch service resending a large system prompt. Through December 31, 2026, that month costs $22.50. From January 1, 2027, the same month costs $45.00:
| Monthly workload | Through Dec 31, 2026 | From Jan 1, 2027 |
|---|---|---|
| 20M input tokens | 20 × $0.75 = $15.00 | 20 × $1.50 = $30.00 |
| 2M output tokens | 2 × $3.75 = $7.50 | 2 × $7.50 = $15.00 |
| Monthly total | $22.50 | $45.00 |
Both dimensions double, so the total doubles exactly — no tier boundary softens it. Cached reads double too ($0.075 → $0.15 per 1M), so a workload leaning on a stable cached prefix sees that line double as well. If you are evaluating Gemini 3.8 Flash on the strength of the $0.75 rate, budget the run at list price and treat the difference as a four-month discount. And do not trust our arithmetic for your actual spend: run your own workload before December 31, reproduce the estimate on the calculator, and verify the per-request line items afterwards with GET /v1/logs, which returns each call’s cost and token breakdown.
What January 1, 2027 does to the rate — an introductory rate stepping up through a cutoff gate. No numbers on purpose: the figures belong in the tables above.
Calling Gemini 3.8 Flash through one OpenAI-compatible base URL
On Modellix’s gateway, Gemini 3.8 Flash is called like any OpenAI-compatible chat model, with one difference: the model field is google/gemini-3.8-flash. The provider/name form is required, and Google models route through Chat Completions or Responses, not the Anthropic Messages endpoint. The base URL is https://llm.modellix.ai/v1, using a single Modellix API key — the same key that reaches the other 27 models:
1 | curl -sS "https://llm.modellix.ai/v1/chat/completions" \ |
If your client already speaks OpenAI’s protocol — an OpenAI SDK, Cursor, Codex, or any tool that takes a base URL and a model name — pointing it at Gemini 3.8 Flash is exactly this one change. For the protocol layer underneath (Chat Completions versus Responses, and where /v1 goes), our OpenAI-compatible API guide walks through it in full.
Gemini Models on the Modellix LLM Gateway
See the current Google model list, protocol routing rules, and pricing notes in the Modellix LLM documentation.
View DocsWhen January 1 arrives: switching is a string change, not a migration
Here is the part Google’s announcement cannot tell you, because it is not a Google feature. When the introductory price expires, the question is not “Gemini got more expensive” — you have known about this since September. The question is who do I switch to, and how much does it cost. Through Google directly, answering that means a new vendor account, a new SDK, re-run evals on another platform. Through a gateway carrying 28 models on one key, it is changing one string in a request you already send.
The stable-alias mechanism is the cleaner version of that. Modellix exposes aliases of the form ~provider/series-latest, and its price page currently maps ~google/gemini-flash-latest → google/gemini-3.8-flash (as of September 4, 2026; the same alias pointed at 3.7 Flash before this release). Alias semantics, from Modellix’s own price page: the alias changes target and your client stays unchanged, and you are billed at the current target’s rate. Call the alias and you never touch code when Google ships the next Flash or reprices this one — the alias moves, your request does not:
1 | curl -sS "https://llm.modellix.ai/v1/chat/completions" \ |
The client code does not change — only the hub’s target does. Illustrative Modellix artwork.
Two honest caveats. First, the alias is a moving target by design: for evaluations and regression tests use a fixed model ID (google/gemini-3.8-flash) so results do not silently change under you — the alias is for production code that should follow the family. Second, “billed at the current target rate” cuts both ways: when the target’s price doubles on January 1, so does your bill, with no code change required. The alias solves the switching problem, not the pricing problem.
For the pricing problem, the answer set is the rest of the catalog. After January 1, 3.8 Flash’s $1.50/$7.50 rate is one that several models on the same key already undercut today — same account, same base URL, same request shape, different model string. Rates below are Modellix billed prices captured September 4, 2026 (USD per 1M tokens, flat-rate unless noted); treat them as a snapshot, not a promise:
| Model (same key, same base URL) | Input / 1M | Output / 1M | Context |
|---|---|---|---|
zai/glm-5.3-flash |
$0.075 | $0.25 | ~1.31M tokens |
qwen/qwen3.8-flash |
$0.16 | $0.47 | 1M tokens |
openai/gpt-5.6-luna (≤272K input tier) |
$0.18 | $1.08 | 1.05M tokens |
deepseek/deepseek-v4-flash |
$0.396 | $1.188 | 1.05M tokens |
anthropic/claude-haiku-4.5 |
$0.90 | $4.50 | 200K tokens |
google/gemini-3.6-flash |
$1.35 | $6.75 | 1.05M tokens |
Source: 28-model snapshot from the Modellix LLM price page, September 4, 2026. Modellix prices and availability change without notice; re-check the live page. This list shows models priced below 3.8 Flash’s post-January rate — not a ranking, and none is claimed cheapest or best for your workload.
One tempting move deserves a direct warning: do not treat Gemini 3.7 Flash as the escape hatch. It carries the same introductory $0.75/$3.75 rate and the same $1.50/$7.50 list price in the Modellix table, so it is likely to cost the same as 3.8 Flash once the introductory period ends; Google has not disclosed a separate expiry for 3.7 in the sources we checked. Its one concrete edge over 3.8 Flash is the 1.05M window noted earlier. Capability differences between the two are Google’s story to tell — their model card and API documentation own the benchmarks — and this page stays on pricing and switching mechanics.
If you would rather choose a gateway than a model, our OpenRouter alternative guide covers what an aggregator actually buys you, and the 28-model LLM API pricing comparison puts the whole catalog on one table for your own before-and-after math.
Why the expiry date matters more than the benchmarks
Google has shipped three Flash versions in six weeks — 3.6, then 3.7, then 3.8 — which is the context here. A model whose price is a four-month promotional window, from a vendor that supersedes its own Flash every few weeks, is not something you re-architect around. It is something you route through an abstraction that makes the next swap as cheap as the first: the alias absorbed the 3.7 → 3.8 transition without a client change and will absorb the next one the same way, while fixed model IDs keep your evaluations reproducible.
On benchmarks we are staying quieter than most pages you will find. Google claims “significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning,” and it has not published numeric deltas against 3.7 Flash — so any article quoting specific 3.7 → 3.8 score gains is filling that gap with its own numbers, and we will not pass those along. Third-party benchmark tables have their place; this page’s job is the pricing window. (Gemini Flash versus Gemini Pro is likewise out of scope — the pricing comparison linked above has both families side by side.)
The summary is short. Through December 31, 2026, Gemini 3.8 Flash is a $0.75/$3.75 model — a cheap way to find out what the newest Flash (as of September 4, 2026) can do on your workload before the price doubles. From January 1, 2027 it is a $1.50/$7.50 model like the rest of its family, and “who do I switch to” is one model string away on the same key and base URL — or zero code changes away if you already call ~google/gemini-flash-latest. The 50% off was never ours to sell you; the ability to leave when the window closes is.
Check Current Gemini 3.8 Flash Rates on Modellix
Log in to see the live price table and per-call cost logs for Gemini 3.8 Flash and the other 27 models on one key.
LoginFrequently Asked Questions
What is Gemini 3.8 Flash API pricing?
As of September 4, 2026, Google’s Gemini 3.8 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens — an introductory rate. From January 1, 2027 those become $1.50 and $7.50 per 1M tokens. The Modellix table mirrors Google’s numbers on this row (struck-through list $1.50/$7.50, billed $0.75/$3.75), with cached reads at $0.075 per 1M during the window.
When does the Gemini 3.8 Flash introductory price expire?
December 31, 2026. Google’s announcement states the introductory price expires that day and that $1.50/1M input and $7.50/1M output apply from January 1, 2027. Re-check live price pages before budgeting — rates change without notice.
Is the $0.75 rate a Modellix discount?
No. The 50% off on Modellix’s price row describes Google’s own schedule: the struck-through $1.50/$7.50 is Google’s list price effective January 1, 2027, and $0.75/$3.75 is Google’s own introductory price. Modellix is a distribution layer that passes Google’s numbers through on this row; it does not discount them.
How do I call Gemini 3.8 Flash through an API?
On Modellix, send a Chat Completions request to https://llm.modellix.ai/v1 with model: google/gemini-3.8-flash and a Modellix API key. Google models use the OpenAI-compatible path, not the Anthropic Messages endpoint; protocol details are in the OpenAI-compatible API guide linked above.
Gemini 3.8 Flash vs 3.7 Flash — which is cheaper?
Pricing-wise they are identical today: both show $0.75/$3.75 introductory in the Modellix table (September 4, 2026) and both carry the same $1.50/$7.50 list price. The one table-level difference is context: 3.8 Flash lists 1M versus 3.7 Flash’s 1.05M. Capability differences belong to Google’s model card, not to a price article.
What is the cheapest Gemini model?
During the window, Gemini 3.8 Flash and 3.7 Flash share the lowest rate in the Gemini Flash line ($0.75/$3.75). After January 1, 2027, as those rows revert toward $1.50/$7.50, Gemini 3.6 Flash’s current $1.35/$6.75 becomes the lowest Gemini-family row on the table, though several non-Google models on the same key undercut all of them. September 4, 2026 snapshot — re-verify before relying on it.
Is Gemini 3.8 Flash free?
No. There is no free tier for Gemini 3.8 Flash through Modellix’s LLM gateway — every successful request there is billed from token usage. On Modellix’s side, the $1 signup credit ended August 19, 2026 — there is no automatic free credit, only pay-as-you-go.
Does the Gemini 3.8 Flash API accept image or audio input?
The model natively supports text, image, video, and audio input, but the Modellix LLM gateway is a text gateway: its documentation instructs text-only prompts and no attached images. For Gemini image generation, the separate media API is covered in our Gemini 2.5 Flash image API guide — a different host, protocol, and billing unit from the text gateway.
Google model pricing and availability change without notice. Figures above were captured September 4, 2026 from Google’s Gemini 3.8 Flash announcement, the Gemini 3.8 Flash model card, and the Modellix LLM price page. The introductory price expires December 31, 2026 — re-verify all rates against the live sources before committing spend. Modellix is an API aggregator with a commercial interest in this page and is not affiliated with Google; Gemini 3.8 Flash is Google’s model, not ours. Access language models from Google, OpenAI, Anthropic, DeepSeek, Qwen, and more through a single API key at modellix.ai.
