Gemini 3.8 Flash API

NewLLM

Gemini 3.8 Flash is the current mainline Gemini Flash route: 1M context, vision, and tool use at the 3.x Flash rate.

Input · USD / 1M tokens$0.45 USDOfficial price~~$0.75 USD~~
Output · USD / 1M tokens$2.25 USDOfficial price~~$3.75 USD~~
Savings40%
Context1M
View tool pricing and billing rules

What are we making today?

Bring an idea, a question, or a first draft.

What this model is for

Gemini 3.8 Flash is the current mainline Gemini Flash route: 1M context, vision, and tool use at the 3.x Flash rate.

Gemini 3.8 Flash is the mainline Flash model in Google's current Gemini family, and the default Gemini entry on OmniaKey for agent work. It has been generally available since September 2, 2026, carries the catalog's fast label, and lists a 1M-token context window with image input and tool use. Google's own page gives 1,048,576 input tokens and 65,536 output tokens. It sits below the Pro models on rate and above the Lite ids.

Three details decide how it behaves in production. Thinking has three levels, low, medium and high, and medium is the default; the minimal setting is not supported and returns an error. Thinking tokens count as output tokens, so a short visible answer can still bill a long reasoning pass. Google also describes implicit caching on eligible prefixes of at least 4,096 tokens, which suits repeated reviews of one document. On the gateway side the rate is a single band, so a large prompt does not step into a second price tier the way the Pro listings do.

That makes 3.8 Flash the starting point for loops that run many times and need a checkable answer each time: classify a ticket, extract fields, pick the next tool, summarise a page. Move to Gemini 4 Argon when one request must hold an entire document set, use video or audio, or answer a question where a missed count is expensive. Gemini 3.7 Flash carries the same listed rate and clients, so a deployment already on that id gains nothing from a swap except the newer snapshot.

Make it the default Gemini workhorse for agent loops that want frontier-adjacent quality at Flash pricing.

Best-fit coding workloads

Bounded agent steps

Run one loop iteration per call: read the tool result, choose the next action, write a two-line reason. Set thinking to low at first, then test medium on the steps that fail, because reasoning tokens count as output tokens. Practical tip: keep the tool schema and system text identical across iterations so the prefix can be cached.

Repeated document review

Put a long policy document or code standard at the front of the prompt and the changing section after it, then ask the same review question many times. Google describes implicit caching on eligible prefixes of 4,096 tokens or more. Practical tip: check the usage fields your client returns instead of assuming a hit from similar-looking text.

High-volume extraction

Pull dates, amounts and party names out of invoices or emails at a volume where the Pro tiers are hard to justify. The single rate band keeps a long attachment from changing the billing logic, and 65,536 output tokens is ample for a batch of small objects. Practical tip: validate each object against a schema and send failures to a second pass.

Hands-on notes

Checked on 2026-10-27 through the OmniaKey API with one standard coding prompt (median of 3 runs) and three capability probes.

Output tokens for our standard coding prompt, reasoning included: 3,013
Cost of that output at current prices: $0.0068
JSON mode: Passed
Tool calling: Passed
Image input: Passed

Technical specification

Catalog facts and integration details verified for this route.

Model ID: gemini-3.8-flash
Context: 1M
Inputs and tools: Vision · Tool use · Fast responses
API endpoint: /v1/chat/completions
Cache hit / 1M: $0.045 USD
Coding clients: Cursor · Cline · Aider

Limits to plan around

The minimal thinking level is not supported and comes back as an error. Validate the level in your client before a run rather than discovering it at the first request of a batch.
A small max_output_tokens value can cut the answer while the model is still reasoning, which yields an empty or partial result that is still billed. Leave headroom and treat a truncated finish reason as a failure.
Google lists Batch, Flex and Priority consumption modes for direct use, and Batch is asynchronous for work that can wait. The gateway catalog shows none of them, so do not build a queue around a mode the endpoint does not offer.

Compare nearby models

Compare workload fit and current listed token rates before choosing a route.

Model	Best for	Context	Input	Output
Gemini 3.8 FlashThis page gemini-3.8-flash	High-volume agent loops, extraction and reviews that need a 1M window at a lower rate tier.	1M	$0.45 USD Official price $0.75 USD	$2.25 USD Official price $3.75 USD
Gemini 4 Argon gemini-4-argon	One large document set per request, video or audio input, or work where a wrong count is costly.	1M	$0.8 USD Official price $2 USD	$4 USD Official price $10 USD
Gemini 3.7 Flash gemini-3.7-flash	Existing deployments already pinned to the 3.7 id, which lists the same rate and capabilities.	1M	$0.45 USD Official price $0.75 USD	$2.25 USD Official price $3.75 USD

SDK quickstarts

Use one base URL and model ID from cURL, Python, or TypeScript.

Streaming chat completion

curl https://api.omniakey.com/v1/chat/completions \
  -H "Authorization: Bearer $OMNIAKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.8-flash",
    "messages": [
      {"role": "user", "content": "Review this pull request"}
    ],
    "stream": true
  }'

Estimate token cost

Enter expected input and output tokens to compare current listed rates.

Input tokensOutput tokens

OmniaKey estimate: $0.09
Official-rate estimate: $0.15
Estimated savings: $0.06

Estimate covers input and output tokens only. Cache usage, retries, tools, and client-side token accounting can change the final charge.

Related coding guides

Deeper selection, setup, and troubleshooting guidance for this model family.

Gemini 3.8 Flash API Pricing and Agent Cost

Google's introductory rate is $0.75 input and $3.75 output per million tokens, but retries and thinking decide the cost of an accepted result.

Read guide

Gemini 4 Release Date: Confirmed, Not Released

The Gemini 4 release date came in stages: Google announced Gemini 4 Argon on September 30, 2026 and opened it to paid API developers on October 13, 2026, then announced Gemini 4 Carbon on October 28, 2026 with no developer date yet.

Read guide

Sources and verification

Model facts and links were checked on 2026-10-27. Provider documentation remains the authority for upstream behavior.

Frequently asked questions

How much does Gemini 3.8 Flash cost on OmniaKey?

OmniaKey charges $0.45 USD per 1M input tokens and $2.25 USD per 1M output tokens, in USD. Official prices are $0.75 USD and $3.75 USD; you save 40%.

What is the context window for Gemini 3.8 Flash?

The listed context window is 1M. Actual usable limits can also depend on the client and upstream request format.

Is Gemini 3.8 Flash cheaper than the Gemini Pro models?

Yes. The catalog places it below every Pro listing, and it has one rate band instead of two, so a long prompt does not jump to a higher band. Gemini 3.5 Flash-Lite and Gemini 2.5 Flash-Lite sit below it. The current rate is in the price table on the model page.

How do thinking tokens affect the bill?

Thinking tokens are billed as output tokens on OmniaKey, so reasoning length counts even when the visible answer is short. The model supports low, medium and high, with medium as the default, and it rejects minimal. Test the levels on your own task set and keep the cheapest one that still passes your checks.

What model id and endpoint should I use?

Send POST requests to https://api.omniakey.com/v1/chat/completions with gemini-3.8-flash in the model field. The body follows the OpenAI-compatible shape, and the same key works for the other Gemini ids if you want to run a comparison. Copy the id exactly, since model strings are case-sensitive.

Can I use Google's Batch or Priority mode through OmniaKey?

The gateway does not list those modes. Google documents Batch, Flex and Priority for direct use on its own API, and Batch is asynchronous with a delayed turnaround. Through the gateway, calls go to the chat completions endpoint, so build any queueing, retry and scheduling logic on your side.

What are the input and output limits?

Google gives this model 1,048,576 input tokens and 65,536 output tokens, and the catalog lists a 1M context window. The output figure includes thinking tokens, so a run that reasons heavily has less room for visible text. Both numbers are ceilings, not a target for a routine request.

Should I switch from Gemini 3.7 Flash?

Both list the same rate, the same clients and the same 1M window, so the reason to move is the newer snapshot and its published documentation, not the bill. Replay a fixed evaluation set against both ids before switching a pinned client, and keep the model string in configuration so you can move back.