Gemini 3.8 Flash is Google’s early-September 2026 workhorse model. It follows Gemini 3.6 Flash and a 3.7 Flash release three weeks before it — Google’s third Flash in roughly six weeks — and is billed as its “best reasoning and coding model yet, at the same speed and low cost of 3.7.”
The design choice behind it is simple: 3.8 Flash works harder. On complex tasks it runs extra reasoning steps and calls tools iteratively, trading some token efficiency for accuracy on long-horizon software engineering, autonomous agents and specialized enterprise work.
Performance
Google published fewer raw numbers for Flash than for its flagship Gemini 4 Argon. What it reports (vendor-reported, September 2026):
| Area | Result |
|---|---|
| DeepSWE v1.1 (long-horizon software engineering) | Outperforms most larger frontier models at a fraction of the cost |
| Vals Finance Agent v2, Harvey’s Legal Agent Benchmark | Beats 3.7 Flash and other frontier models |
| HLE-Verified | 54.9% across STEM, humanities and professional fields |
| Glean document-heavy workflows | More than 3x as many tasks completed as 3.7 Flash |
| Prompt-injection robustness (Gray Swan) | “Significant leap” over earlier Gemini models |
Showcase builds from Google include a DOS-style Google Maps clone in a single prompt, a 3D wizard-castle game made through a looping instruction in Antigravity, and an interactive 3D hardware teardown visualizer in AI Studio.
Specs
| Spec | Value |
|---|---|
| API model ID | gemini-3.8-flash |
| Input | Text, image, video, audio, PDF |
| Output | Text |
| Input token limit | 1,048,576 |
| Output token limit | 65,536 |
| Thinking | Low, medium, high (minimal returns an error) |
| Tools | Function calling, code execution, Search and Maps grounding, URL context, file search, structured outputs, computer use (preview) |
| Serving options | Standard, Batch API, Flex and Priority inference, context caching |
| Not supported | Audio or image generation, Live API |
Pricing
| Item (per 1M tokens) | Through Dec 31, 2026 | From Jan 1, 2027 |
|---|---|---|
| Input | $0.75 | $1.50 |
| Output (incl. thinking) | $3.75 | $7.50 |
| Context caching | $0.075 | $0.15 |
| Cache storage (per hour) | $0.50 | $1.00 |
Grounding with Google Search or Maps includes 5,000 free requests per month shared across Gemini 3 and later models, then $14 per 1,000. There is a free tier in AI Studio and the API, and paid-tier content is not used to improve Google’s products.
Gemini 3.8 Flash Cyber
Alongside Flash, Google released 3.8 Flash Cyber, a variant tuned for defensive security and gated behind the new Fairwind Program. Google reports a CWE-Bench pass@1 of 47.2% (against 47.8% for a leading frontier model at far higher cost), a success rate above 70% on its internal 20-language vulnerability benchmark, and 2.6x more correct Chrome patches than larger commercial models.
Availability
- Developers: Gemini API and AI Studio, Google Antigravity, Android Studio and Stitch.
- Enterprises: Gemini Enterprise.
- Consumers: Gemini app, AI Mode in Search and Gemini in Sheets for Google AI Pro and Ultra subscribers.
- Sibling models in September: Gemini 3.8 Live and 3.8 Live Extended Thinking (voice), plus Gemini 3.8 Flash TTS and Flash-Lite TTS (speech generation).
How 3.8 Flash compares
- vs Gemini 3.6 Flash: a clear step up on coding, agentic and finance/legal evals at the same introductory price tier; both double to $1.50 / $7.50 in 2027.
- vs Gemini 4 Argon: Argon is the frontier model, gated for now; 3.8 Flash is the model you can build on today.
- vs Claude Haiku 5.5: Haiku is far cheaper ($0.10 / $0.50) for narrow, high-volume tasks; Flash is stronger and multimodal for agentic work.
- vs GPT-6.1 Sol: Sol costs $2 / $10 and targets near-Astra intelligence; Flash is a cheaper tier for work that does not need frontier peaks.
Limitations
- Token use. The “works harder” design can consume more tokens at high effort; Google itself recommends lower effort or 3.7 Flash for efficiency-first use.
- Price change. Budgets built on $0.75 / $3.75 need to plan for the 2027 doubling.
- Few published numbers. Beyond HLE-Verified, most claims are relative (“outperforms most larger models”) — wait for third-party evaluations.
For the wider picture, see the 2026 LLM guide.
Best for
- Long-running coding agents and autonomous workflows at scale (Antigravity, AI Studio)
- Finance, legal and document-heavy enterprise agents (Vals Finance Agent v2, Harvey Legal Agent Benchmark)
- Multimodal understanding over video, audio, images and PDFs
- Cost-sensitive replacement for frontier flagships on everyday agentic tasks
Pros & cons
Strengths
- Google's most intelligent Flash model — substantial gains over 3.7 Flash on software engineering, agentic tasks and multi-step reasoning, with Flash latency and price
- Priced at $0.75 / $3.75 per 1M through 2026, a fraction of frontier flagships ($2–$10 input) while Google says it beats most larger models on DeepSWE v1.1
- Truly multimodal in — text, image, video, audio and PDF — with 1,048,576 input tokens and 65,536 output tokens
- Wide availability from day one — Gemini API, AI Studio, Antigravity, Android Studio, Gemini app and AI Mode, Gemini Enterprise
Limitations
- Works harder by design — more reasoning steps and tool calls can mean more tokens at higher effort; Google suggests lower effort or staying on 3.7 Flash for efficiency-first work
- Introductory price doubles to $1.50 / $7.50 on January 1, 2027
- Thinking supports only low, medium and high — minimal is rejected with an error
- Text-only output (no image or audio generation) and no Live API; Google ships separate Live and TTS variants
Frequently asked questions
When was Gemini 3.8 Flash released?
Around September 2, 2026, as Google's third Flash release in about six weeks — after 3.6 Flash (July 21) and 3.7 Flash — together with the restricted-access Gemini 3.8 Flash Cyber. Related audio models, Gemini 3.8 Live and 3.8 Flash TTS, followed in September.
How much does Gemini 3.8 Flash cost?
$0.75 per 1M input tokens and $3.75 per 1M output tokens (including thinking tokens) through December 31, 2026, rising to $1.50 / $7.50 from January 1, 2027. Cached input is $0.075 per 1M (then $0.15).
What is the model ID and context window?
`gemini-3.8-flash` with 1,048,576 input tokens and 65,536 output tokens. Inputs can be text, image, video, audio and PDF; output is text. Thinking levels are low, medium and high.
How does Gemini 3.8 Flash compare with Gemini 4 Argon?
Argon is Google's frontier model announced September 30 with a $2 / $10 introductory price, but it is limited to trusted defenders for now. 3.8 Flash is generally available, cheaper and faster, and sits below Argon in capability.
What is Gemini 3.8 Flash Cyber?
A cybersecurity variant built on the same core, available only to trusted defenders through Google's Fairwind Program. Google reports a CWE-Bench pass@1 of 47.2% (vs 47.8% for a leading frontier model at a much higher cost) and 2.6x more correct Chrome vulnerability patches than larger commercial models.