Explore ProductsAI WorkspaceEnterpriseApps & SystemsManaged agentsWhite labelAPI
PlatformsDesktop AppChrome ExtensionWeb App
Explore ResourcesResourcesUse casesCase studies
Mission
Get started
SPIRITT logoSPIRITTGoogle DeepMindGemini 3.8 Flash

Gemini 3.8 Flash brings frontier agent work up to speed

Google's September 2 workhorse pairs a 59 Artificial Analysis Intelligence score with 304.6 output tokens per second, 1M multimodal context, and frontier-adjacent coding. Gemini 3.8 Flash is now available in SPIRITT Workspaces for agentic apps and workflows.

Explore SPIRITT Workspaces
See how it works

Gemini 3.8 Flash across X

The SPIRITT availability update, official release, independent speed and intelligence evidence, launch reactions, and a useful warning about benchmark optimism.

A faster Flash model for coding and complex knowledge work

The independent speed result is exceptional. The current official table also shows both strong agent results and visible limits.

What Google launched, and what you can build in SPIRITT

Google positions Gemini 3.8 Flash as its production workhorse for software engineering and agentic knowledge workflows. It builds on Gemini 3.7 Flash, accepts text, image, audio, and video, and can return up to 64K text tokens from a 1M-token input context.

Artificial Analysis measured the high-effort setting at 59 on its Intelligence Index and 304.6 output tokens per second. Google's current model card reports 73.7% on DeepSWE v1.1 and 89.4% on Terminal-Bench 2.1, while Terminal-Bench 4.0 and OSWorld show where stronger frontier systems still lead.

Introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Higher effort can improve quality, but it uses more reasoning tokens and increases time to first token.

The exact Gemini 3.8 Flash route is now available in SPIRITT Workspaces and passed a live completion check on September 2. Select it in the model picker to build agentic apps and workflows with the new release.

1M context64K outputText + image + audio + videoConfigurable effort$0.75 / $3.75 introAvailable in SPIRITT

AA Intelligence Index v4.1.1

Claude Fable 5.1 max
66
Claude Opus 5 max
63
Claude Fable 5 max
62
Grok 4.6 high
61
Gemini 3.8 Flash high
59
Gemini 3.7 Flash high
56

Artificial Analysis release snapshot, September 2, 2026. Gemini 3.8 Flash high scores 59, up from Gemini 3.7 Flash high at 56, but behind the displayed frontier leaders.

Where Gemini 3.8 Flash is fast, strong, and still uneven

Independent measurements establish overall speed, intelligence, and cost. Google's model card then supplies capability-specific comparisons, including the losses.

Independent speed
Winner

Output tokens / second

Gemini 3.8 Flash high
304.6
Gemini 3.7 Flash high
284.8
Muse Spark 1.2
117
GPT-5.6 Terra max
108
Claude Fable 5.1 max
66

Artificial Analysis release snapshot. Gemini 3.8 Flash was the fastest model in its tracked set at publication.

Independent effort cost

AA cost per Index task

Gemini 3.8 low
$0.24
Gemini 3.8 medium
$0.41
Gemini 3.8 high
$0.58

Artificial Analysis estimates. High effort raises measured intelligence, cost, and latency relative to lower settings.

Vendor-compiled coding
Near frontier

DeepSWE v1.1

Claude Opus 5 max
74%
Gemini 3.8 Flash
73.7%
GPT-5.6 Sol max
72.7%
Gemini 3.7 Flash
65.3%

Google's current model card reports 73.7%, superseding the 71.0% in the launch-day graphic. Comparator provenance follows Google's methodology.

Vendor-compiled terminal agents
Winner

Terminal-Bench 2.1

Gemini 3.8 Flash
89.4%
Claude Opus 5 max
89.1%
GPT-5.6 Sol max
88.8%
Gemini 3.7 Flash
85.8%

Google model card. Gemini results are self-computed; comparator provenance follows Google's methodology.

Vendor-compiled finance agents
Winner

Finance Agent v0.2

Gemini 3.8 Flash
61.4%
Gemini 3.7 Flash
59%
Claude Opus 5 max
58.6%
GPT-5.6 Terra max
54.4%

Google model card. Higher is better; this is not an independent rerun by SPIRITT.

Vendor-compiled multimodal
Winner

CharXiv reasoning

Gemini 3.8 Flash
86.2%
GPT-5.6 Terra max
85.9%
GPT-5.6 Sol max
85.8%
Gemini 3.7 Flash
84.5%

Google model card. Differences among the leading displayed systems are narrow.

Harder agent benchmark

Terminal-Bench 4.0

Claude Opus 5 max
51.8%
GPT-5.6 Sol max
37.3%
GPT-5.6 Terra max
23.6%
Gemini 3.8 Flash
19.1%
Gemini 3.7 Flash
11.2%

An important limitation: Gemini 3.8 improves over 3.7 but trails the strongest displayed comparators on the newer benchmark version.

Vendor-compiled computer use

OSWorld-Verified

Claude Opus 5 max
75.4%
GPT-5.6 Sol max
62.6%
Gemini 3.8 Flash
59%
Gemini 3.7 Flash
50.6%

Google model card. Stronger than Gemini 3.7 Flash, but not the leading displayed computer-use result.

Independent source: Artificial Analysis Gemini 3.8 Flash release and live model pages, observed September 2, 2026. Vendor source: Google's current Gemini 3.8 Flash model card and linked evaluation methodology. Scores, prices, and rosters can change. Run repeated evaluations on your own work before switching a production route.

How It Works

Choose Gemini 3.8 Flash in the picker, then build the agentic workflow

01

Open a workspace

Open a workspace and land in a fully equipped cloud computer: browser, files, terminal, integrations, and memory. No local setup. No thin chat box pretending to be an agent.

Create a workspace in SPIRITT
02

Choose Gemini 3.8 Flash

Open the model picker and select Gemini 3.8 Flash. The exact route is live in SPIRITT Workspaces and has passed a real completion check.

Gemini visual for the Gemini 3.8 Flash model-picker step
03

Build or automate

Tell it what to ship or what to run. Gemini 3.8 Flash can code, call tools, drive the browser, coordinate multi-step work, and keep going while you step away. The point is not another chat window. It is an agentic environment where Gemini 3.8 Flash actually does the job.

Build an agentic app in SPIRITT

Build with Gemini 3.8 Flash now.

Open a SPIRITT Workspace, choose Gemini 3.8 Flash in the model picker, and turn the release into a working agentic app or workflow.

Open SPIRITT
Google model card
Independent evaluation
Official launch
Questions

Gemini 3.8 Flash FAQ

01What is Gemini 3.8 Flash?+
Google DeepMind's September 2026 production workhorse for fast software engineering, agentic knowledge work, and multimodal analysis. It builds on Gemini 3.7 Flash and supports configurable effort levels.
02How fast is Gemini 3.8 Flash?+
Artificial Analysis measured 304.6 output tokens per second at high effort and called it the fastest model in its tracked set at publication. High effort still had a 13.39-second time to first token, so not every request begins instantly.
03What do the effort settings change?+
They trade cost and latency for measured quality. Artificial Analysis reported Index scores of 51, 57, and 59 at low, medium, and high effort, with estimated task costs of $0.24, $0.41, and $0.58 respectively.
04What are the main limitations?+
Google warns that the model can hallucinate, misunderstand complex instructions, miss information in very long prompts, respond slowly, or rarely time out. The model card also shows that Terminal-Bench 4.0 and OSWorld remain weaker than the strongest displayed frontier comparators.
05How much does Gemini 3.8 Flash cost?+
Google lists introductory pricing of $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Production budgets should use current official pricing after that date.
06What can fit in the context window?+
The model card lists a 1,048,576-token input window and up to 65,536 text output tokens. It accepts text, images, audio, and video, but Google still warns that important details can be missed in very long prompts.
07Is Gemini 3.8 Flash available in SPIRITT Workspaces?+
Yes. The exact Gemini 3.8 Flash route appeared in SPIRITT's authenticated model roster on September 2 and passed a live completion check. Open the model picker to use it in an agentic app or workflow.
Buy from builders who use what they sellBuilt usingSPIRITT