Hassann

Posted on Sep 3 Originally published at apidog.com

What is Gemini 3.8 Flash?

Gemini 3.8 Flash: More Agentic Reasoning at the Same Token Price

Gemini 3.8 Flash is Google’s newest Flash-tier model, released on September 2, 2026, alongside the gated security model Gemini 3.8 Flash Cyber. It is the third Flash release in six weeks, following Gemini 3.6 Flash on July 21 and Gemini 3.7 Flash on August 13. Introductory pricing matches 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Google describes it as “our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows.”

Try Apidog today

The main change is behavior, not price or context size. Gemini 3.8 Flash takes smaller reasoning steps, verifies its work, and calls tools iteratively. This improves agentic benchmark results, but also increases token usage by design. Artificial Analysis measured approximately 30% more output tokens than Gemini 3.7 Flash.

So the per-token price is unchanged, but the per-task cost can be higher.

Gemini 3.8 Flash at a glance

Specification	Value
API model ID	`gemini-3.8-flash`
Released	September 2, 2026
Based on	Gemini 3.7 Flash, according to the DeepMind model card
Context window	1,048,576 input tokens
Maximum output	65,536 tokens
Input modalities	Text, image, video, audio, PDF
Output modalities	Text only
Thinking levels	`low`, `medium` (default), `high`; `minimal` returns an error
Introductory pricing	$0.75 input / $3.75 output per million tokens through December 31, 2026
Standard pricing	$1.50 input / $7.50 output per million tokens from January 1, 2027
Context caching	$0.075 per million cached tokens introductory; $0.15 standard
Batch API	50% off: $0.375 input / $1.875 output introductory
Knowledge cutoff	March 2026
Developer availability	Gemini API, Google AI Studio, Android Studio, Google Antigravity, Stitch, Gemini Enterprise
Consumer availability	Gemini app for AI Pro and Ultra subscribers, AI Mode in Search, Gemini in Google Sheets
Gemini 3.7 Flash	Fully supported; no deprecation date announced

Three details matter during migration:

The default thinking level is medium, not high as with Gemini 3 Pro.
minimal returns a validation error instead of falling back to low. See the thinking levels guide.
Although the stated knowledge cutoff is March 2026, the model card notes that some domains may have knowledge limited to January 2025.

What Gemini 3.8 Flash is

Flash is Google’s workhorse tier: it is intended for most production traffic, while Pro targets the hardest problems. Google calls Gemini 3.8 Flash its “most intelligent workhorse model yet,” but the DeepMind model card describes it as based on Gemini 3.7 Flash.

A practical description is therefore: Gemini 3.7 Flash refined for long-running software engineering and agent workflows.

The release cadence is unusually fast:

Gemini 3.6 Flash: July 21, 2026, at $1.50 / $7.50
Gemini 3.7 Flash: August 13, 2026, at the $0.75 / $3.75 introductory rate
Gemini 3.8 Flash: September 2, 2026, at the same introductory rate

Google calls this the third Flash release in six weeks. Artificial Analysis counts a fourth Flash model in under four months because its tally includes Gemini 3.5 Flash-Lite. No Gemini 3.8 Pro, Gemini 4, or new Flash-Lite model shipped with this release.

Gemini 3.7 Flash remains supported, and Google has not announced a deprecation date or a timeline for a Pro update. The previous Gemini 3.7 Flash what’s-new post is already describing the previous generation only three weeks after publication.

The “works harder” design—and its cost

Google describes Gemini 3.8 Flash’s changes in concrete terms. On difficult tasks, it:

Executes additional reasoning steps
Breaks work into smaller reasoning steps
Verifies intermediate results
Calls tools iteratively

Google also warns that the model can use more tokens on long-running and complex tasks, especially at higher effort levels.

Artificial Analysis measured the impact using its Intelligence Index:

Thinking level	Cost per task	Time per task
High	$0.58	2.5 minutes
Medium	$0.41	—
Low	$0.24	0.8 minutes

At high reasoning, Gemini 3.8 Flash averaged 48,000 output tokens per task—about 30% more than Gemini 3.7 Flash. The comparable 3.7 Flash high-reasoning task cost was $0.40 and took 2.2 minutes.

This makes thinking_level a routing decision rather than a global configuration:

Use low for latency-sensitive endpoints.
Use medium for most agent workflows.
Use high when task completion matters more than token or time budgets.

Speed is otherwise similar to Gemini 3.7 Flash. Logan Kilpatrick of Google described it as “same price as 3.7, [and] ~the same speed.” Artificial Analysis measured 302.1 output tokens per second at high reasoning. Its 13.30-second time to first token reflects model-side thinking, not necessarily network latency.

See the pricing breakdown for examples across 1,000 agent tasks per day.

What the benchmarks say

Google published the following benchmark results on its DeepMind Flash page:

Benchmark	Gemini 3.8 Flash	Gemini 3.7 Flash	Claude Opus 5	GPT-5.6 Sol	GPT-5.6 Terra	Claude Sonnet 5
Vals Finance Agent v2	61.4%	59.0%	58.6%	53.8%	54.4%	53.9%
Harvey Legal Agent Benchmark	10.0%	8.8%	6.7%	2.5%	0.8%	5.0%
HLE-Verified	54.9%	53.6%	54.4%	54.5%	51.1%	31.0%

The gains over Gemini 3.7 Flash are 2.4, 1.2, and 1.3 percentage points. The finance and legal results place a Flash-priced model ahead of Opus 5 and GPT-5.6 Sol on these published tests.

Claude Fable 5.1, released the day before, is not included in Google’s tables. The three-way comparison uses Anthropic’s Opus 5 and Sonnet 5 results as the closest published entries.

Google also reports:

On DeepSWE v1.1, Gemini 3.8 Flash outperforms most larger frontier models at a fraction of the cost. The model card includes image-based percentage tables; Google reports Gemini 3.7 Flash at 65.3%.
On document-heavy, long-running workflows, an internal evaluation completed more than three times as many tasks as Gemini 3.7 Flash.
On the Gray Swan prompt-injection evaluation, it achieved a “significant leap,” without a published figure.

Text results for SWE-Bench Pro, Terminal-bench, and OSWorld were not published for Gemini 3.8 Flash.

Artificial Analysis independently scored Gemini 3.8 Flash high at 59 on its Intelligence Index, compared with:

Gemini 3.7 Flash: 56
Gemini 3.6 Flash: 52
GPT-5.6 Sol xhigh: 59
Grok 4.6 medium: 59
GPT-5.6 Terra max: 57
Claude Fable 5.1 medium: 57
Muse Spark 1.2 xhigh: 57

On the τ³-Banking tool-use evaluation, Gemini 3.8 Flash scored 45%, 12 points above Gemini 3.7 Flash. The 3.8 versus 3.7 comparison weighs these gains against the added token consumption.

Pricing: same per token, more per task

The official pricing page lists Gemini 3.8 Flash alongside Gemini 3.6 and 3.7 Flash. Prices double on January 1, 2027.

Item	Through December 31, 2026	From January 1, 2027
Input	$0.75 / 1M tokens	$1.50 / 1M tokens
Output, including thinking	$3.75 / 1M tokens	$7.50 / 1M tokens
Cached input	$0.075 / 1M tokens	$0.15 / 1M tokens
Cache storage	$0.50 / 1M tokens / hour	$1.00 / 1M tokens / hour
Batch input / output	$0.375 / $1.875	$0.75 / $3.75

Two billing details are easy to miss:

Thinking tokens are billed as output tokens and reported separately as usageMetadata.thoughtsTokenCount. A short high-reasoning answer can cost more than a longer low-reasoning answer.
Google Search grounding has a separate meter: 5,000 free requests per month shared across Gemini 3.x models, then $14 per 1,000 requests.

AI Studio and the API provide a rate-limited free tier. Google states that free-tier data is used to improve its products. Per-model rate limits are shown in AI Studio.

Billing tiers unlock as follows:

Tier 1: link a billing account
Tier 2: spend $100 and wait three days
Tier 3: spend $1,000 and wait 30 days

The free-access guide explains the free path, while the Batch API guide covers the 50% discount for jobs that can wait.

How to call Gemini 3.8 Flash

Google now treats the Interactions API as the primary API for Gemini 3.x. generateContent is considered legacy but remains fully supported with no announced sunset date.

A minimal Interactions API request looks like this:

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-[REDACTED CREDENTIAL] \
  -H 'Content-Type: application/json' \
  -d '{"model":"gemini-3.8-flash","input":"Explain HTTP caching in 3 sentences.","generation_config":{"thinking_level":"medium"}}'

For multi-turn conversations, send previous_interaction_id so the server can maintain state.

Function calling uses a JSON Schema tool declaration. The model returns a function_call step, and your application sends back a function_result containing both call_id and name. Both fields are required for Gemini 3.8 Flash. In legacy generateContent, the corresponding field is id.

The API walkthrough includes REST and Python examples. The function calling guide explains iterative tool loops and how to cap them.

When migrating from Gemini 3.7 Flash, watch for these incompatibilities:

minimal thinking is rejected.
thinking_budget is replaced by thinking_level.
candidate_count is unsupported.
Google recommends leaving temperature at its default of 1.0; lowering it may cause looping or degraded performance.

Use the migration guide for before-and-after JSON examples.

What Gemini 3.8 Flash cannot do

Gemini 3.8 Flash does not support:

Audio generation
The Live API
Image generation
Image segmentation, which is not supported on Gemini 3 models

Although the model accepts text, images, video, audio, and PDFs as input, its output is text only. For real-time voice or image output, use Gemini 3.8 Flash as the reasoning and tool-calling layer alongside other services.

For inexpensive, high-throughput text generation without heavy reasoning, Gemini 3.5 Flash-Lite remains available at $0.30 input / $2.50 output per million tokens.

Supported features include:

Function calling
Structured outputs
Context caching
Code execution
Search and Maps grounding
Batch API
Computer use in preview

Gemini 3.8 Flash Cyber: the gated twin

Gemini 3.8 Flash Cyber shipped on the same day, but it is not publicly available. Access is limited to Google’s Fairwind Program for trusted government authorities, critical infrastructure operators, and software maintainers.

The program requires background checks and controls such as phishing-resistant MFA. There is:

No public API
No public pricing
No self-hosting option

Gemini 3.8 Flash Cyber replaces the Gemini 3.5 Flash Cyber limited pilot.

Google reports a vulnerability-discovery success rate above 70% across 20 programming languages. Google’s Chrome security team also reports that Cyber produced 2.6 times more correct vulnerability patches than the best much-larger commercial models. Both figures are Google-reported.

The Cyber explainer covers eligibility, obligations, and the implications for teams without access.

Test Gemini 3.8 Flash in Apidog

Because Gemini 3.8 Flash’s cost is driven by task-level token usage, a successful HTTP response is not enough. Test how many tokens each request consumes.

In Apidog:

Store GEMINI_API_KEY as an environment variable.
Save both Interactions API and generateContent requests as endpoints.
Assert that the response status is 200.
Validate the JSON fields consumed by your application.
Set a ceiling for usageMetadata.thoughtsTokenCount.
Run the same prompt at low, medium, and high.
Compare token usage and latency for your own workload.

Streaming responses use SSE. Set includeThoughts: true when debugging thought summaries, then follow the SSE testing guide.

Schedule the scenario daily so a token regression fails an assertion before it reaches your invoice. You can download Apidog and set up the test on the free plan.

FAQ

Is Gemini 3.8 Flash a new model or an update to Gemini 3.7 Flash?

The DeepMind model card says it is based on Gemini 3.7 Flash. Treat it as a tuned successor with the same price, speed, and context window, but more reasoning and tool-calling steps on difficult tasks. Gemini 3.7 Flash remains fully supported with no announced deprecation date.

Why does Gemini 3.8 Flash use more tokens than Gemini 3.7 Flash?

That behavior is intentional. Google says Gemini 3.8 Flash can use more tokens on long-running and complex tasks. Artificial Analysis measured approximately 30% more output tokens in its benchmark.

Set thinking_level to medium or low for routes that do not need maximum reasoning. The thinking levels guide includes per-level cost examples.

What is the Gemini 3.8 Flash context window?

The model supports 1,048,576 input tokens and 65,536 output tokens. Context caching costs $0.075 per million cached tokens through December 31, 2026.

Can I use Gemini 3.8 Flash in the free Gemini app?

Launch coverage lists the Gemini app for Google AI Pro and Ultra subscribers, not the free app tier. Developers can use a rate-limited free tier in AI Studio and the API.

How does Gemini 3.8 Flash compare with Claude Fable 5.1?

Artificial Analysis scores Gemini 3.8 Flash at 59 and Claude Fable 5.1 at 57. Claude Fable 5.1 costs $10 input / $50 output per million tokens—approximately 13 times Gemini 3.8 Flash’s introductory rate for both token types.

That gap narrows when you account for Gemini 3.8 Flash’s higher token consumption per task. See the Claude Fable 5.1 overview.

Conclusion

Gemini 3.8 Flash is a workhorse that reasons longer. The trade-off is explicit: modest gains on agentic benchmarks and a 12-point improvement on tool use, offset by approximately 30% more output tokens at high reasoning.

If Gemini 3.7 Flash was hitting limits on complex agent tasks, Gemini 3.8 Flash offers the upgrade at the same per-token price. If your workload is latency-sensitive chat, use low or stay on Gemini 3.7 Flash.

Start with the API walkthrough, send a request from Apidog with a token-count assertion, and choose the thinking level for each route based on your own workload—not only the launch benchmarks.

DEV Community