Guide
Gemini 3.8 Flash on OpenClaw and Hermes
Google lists gemini-3.8-flash as a stable model and describes it as its most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows. That is a Flash-tier model aimed squarely at the kind of work an always-on agent does. Here is how to run it on OpenClaw or Hermes Agent, and what is actually known about it.
What Gemini 3.8 Flash Is
Flash has always been Google's throughput tier — the model you point at work that runs constantly and has to stay cheap. The 3.8 generation keeps that position but is explicitly pitched at long-horizon tasks rather than short ones, which is the part that matters for agents. An agent handling a multi-step job holds a growing context and makes many tool calls; a model tuned for that profile behaves very differently from one tuned for single-turn chat.
Where It Sits in the Flash Line
| Model ID | Position |
|---|---|
gemini-3.8-flash | Newest. Positioned for long-horizon software engineering, autonomous agents and complex enterprise workflows. |
gemini-3.7-flash | Previous Flash generation, still stable and selectable. |
gemini-3.6-flash | Widely deployed. Covered in its own guide. |
gemini-3.5-flash | Older Flash tier, plus a flash-lite variant. |
Google has not published a context window, knowledge cutoff, or per-token price for 3.8 Flash on its model list. Rather than guess at those here, check the Gemini API model page and Google's pricing page before you budget a heavy workload against it. Treat any blog post quoting exact 3.8 Flash pricing with suspicion until it cites Google directly.
How to Run It on OpenClaw Launch
Both routes work with a bring-your-own-key setup — a Google AI Studio key for the Gemini API directly, or an OpenRouter key if you would rather keep one provider account for everything.
- Deploy an agent from openclawlaunch.com — about 30 seconds. Bring-your-own-key model selection needs a paid plan; the free trial runs a fixed set of free models, so use it to check the product rather than this model.
- In your dashboard, add your Google AI Studio key (or an OpenRouter key) under BYOK settings.
- Open the model picker and type gemini 3.8. Searching reveals the live provider catalog rather than only the curated shortlist, so the model appears as soon as your provider lists it — no redeploy and no config file edit.
- Select it. The change applies to the running agent immediately.
The same flow covers Hermes Agent: deploy a Hermes instance instead, add the key, and pick the model. See Hermes + Gemini for the Hermes-specific notes.
Self-Hosted Configuration
If you run your own instance, set the model id gemini-3.8-flash against the Gemini provider, or google/gemini-3.8-flash if you are routing through OpenRouter. The provider setup itself is unchanged from earlier Gemini generations, so the existing OpenClaw + Gemini guide covers the credentials and config file layout — only the model string differs.
Because it is a current stable model rather than a preview, it does not need a preview-channel flag or an allowlist request. If your instance cannot see it, the usual cause is a provider account that has not been enabled for the newest models rather than anything in the agent config.
When to Pick It
- Pick 3.8 Flash for agents that run all day on many small tool calls, and for multi-step coding or workflow jobs where the session grows long. That is the profile Google is explicitly targeting.
- Stay on 3.6 Flash if you have a tuned, working setup and no complaints — there is no reason to chase a version number. See the 3.6 Flash guide.
- Reach for a frontier model — Claude Fable 5.1, GPT-6 Astra — for hard one-off reasoning and code review, and keep Flash as the everyday driver. Switching per task takes one command.
/model chat command, so the cheap way to evaluate 3.8 Flash is to run it as your driver for a few days and keep a frontier model for reviews. See multi-model switching.