The short version
Gemini 3.8 Flash is the next step in Google's Gemini 3 model family, built directly on Gemini 3.7 Flash[1]. The headline is value: it closes most of the software-engineering gap to far more expensive flagship models — Claude Opus 5 and GPT-5.6 Sol — while staying in the low-cost "Flash" price tier[1][2]. For anyone running agents or coding workflows at volume, that combination is the whole story.
What Gemini 3.8 Flash is
Gemini 3.8 Flash accepts text, images, audio, and video, with a context window of up to 1M tokens and a 64K-token output[1]. It keeps the customizable "effort levels" from the 3.x line, so you can dial the mix of quality, cost, and latency to the task[1]. Google positions it as the model for "cost-effective scaling of general-purpose, production-ready agents," targeting software engineering, agent tasks, and complex knowledge workflows[1].
Its knowledge cutoff is March 2026 (with some domains still limited to January 2025, consistent with the Gemini 3 family)[1].
Price and where to get it
Gemini 3.8 Flash launches at the same introductory price as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens[2]. The model card lists the non-cached regular price at $1.50 input and $7.50 output[1]. That makes it dramatically cheaper than the flagship competition at list price.
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Gemini 3.8 Flash | $0.75 ($1.50) | $3.75 ($7.50) |
| Claude Sonnet 5 | $2.00 | $10.00 |
| GPT-5.6 Terra | $2.00 | $12.00 |
| GPT-5.6 Sol | $4.00 | $20.00 |
| Claude Opus 5 | $5.00 | $25.00 |
List prices from Google's Gemini 3.8 Flash model card, as of September 2026[1].
It ships across the Gemini app, Google AI Studio, the Gemini API, Google Antigravity, Google AI Mode, and the Gemini Enterprise Agent Platform[1]. Google marked it generally available in the Gemini API in mid-September[3].
Benchmarks vs the field
On long-horizon software engineering (DeepSWE v1.1), 3.8 Flash scores 73.7%, up from 65.3% for 3.7 Flash — within a point of Claude Opus 5 (74.0%) and ahead of GPT-5.6 Sol (72.7%)[1]. On agentic terminal coding (Terminal-bench 2.1) it leads the table at 89.4%, narrowly ahead of Opus 5 (89.1%) and GPT-5.6 Sol (88.8%)[1].
| Benchmark | 3.8 Flash | 3.7 Flash | Claude Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| DeepSWE v1.1 (software eng.) | 73.7% | 65.3% | 74.0% | 72.7% |
| Terminal-bench 2.1 (agentic coding) | 89.4% | 85.8% | 89.1% | 88.8% |
| Vals Finance Agent v2 | 61.4% | 59.0% | 58.6% | 53.8% |
| Harvey's Legal Agent | 10.0% | 8.8% | 6.7% | 2.5% |
| OSWorld-2.0 (computer use) | 59.0% | 50.6% | 75.4% | 62.6% |
| HLE-Verified (expert reasoning) | 54.9% | 53.6% | 54.4% | 54.5% |
Selected results from Google DeepMind's Gemini 3.8 Flash model card, September 2026[1].
The pattern is consistent: 3.8 Flash is competitive with the two expensive flagships on coding and agentic-terminal work, and it even leads on the finance-agent and legal-agent tasks shown here. The one area where Opus 5 still clearly wins is computer use (OSWorld-2.0, 75.4% vs 59.0%)[1].
Gemini 3.8 Flash Cyber
Alongside the main model, Google launched Gemini 3.8 Flash Cyber, which it calls its "most capable cybersecurity model" with frontier-level performance in vulnerability detection and automated patching[2]. Unlike the broadly available Flash model, Cyber is gated: it's distributed to "trusted defenders" through Google's new Fairwind Program[2]. Both models share the same foundational intelligence, tuned for different deployment environments[2].
What it means for small business & creators
- Cheaper agent workloads. If you run coding agents, summarization, or data tasks through the Gemini API, 3.8 Flash delivers near-flagship coding results at a fraction of Opus 5 or GPT-5.6 Sol pricing[1].
- No app-side decision to make. The model is simply available in the Gemini app, AI Studio, and the API — no new plan or contract is required to try it[1].
- Longer documents, fewer chunks. The 1M-token window means large PDFs, codebases, or transcripts can be processed with less manual splitting[1].
- Cyber is not for you (probably). Flash Cyber is aimed at security teams and researchers, not general business users, and access is restricted via Fairwind[2].
The caveats
This is Google's fourth "Flash" release in a short window, and the pace means version numbers are moving fast — check which model your API calls are actually pinned to[3]. The model card also notes the usual foundation-model limits (possible hallucinations, occasional slowness or timeouts) and flags that at higher effort levels the model may use more tokens to maximize performance[1]. Finally, if your workflow is heavily computer-use dependent, Opus 5 still holds a real edge[1].
Bottom line
Verdict: Worth testing this week if you pay for any Gemini tier or use the Gemini API. Gemini 3.8 Flash is the clearest "more capability for the same money" update Google has shipped in a while — strong on coding and agents, at a price the flagships can't match.
Frequently asked questions
Is Gemini 3.8 Flash more expensive than 3.7 Flash?
No — it launches at the same introductory price of $0.75 per million input tokens and $3.75 per million output tokens[2].
What's the difference between 3.8 Flash and 3.8 Flash Cyber?
Both share the same underlying intelligence, but Flash Cyber is tuned for cybersecurity — vulnerability detection and automated patching — and is only available to trusted defenders through the Fairwind Program[2].
Where can I use Gemini 3.8 Flash?
In the Gemini app, Google AI Studio, the Gemini API, Google Antigravity, Google AI Mode, and the Gemini Enterprise Agent Platform[1].
Sources
- Google DeepMind — Gemini 3.8 Flash model card — deepmind.google
- Google — "Introducing Gemini 3.8 Flash and 3.8 Flash Cyber" — blog.google
- Google AI for Developers — Gemini API changelog — ai.google.dev