For a new OpenClaw setup, test Gemini 3.8 Flash for everyday work, Gemini 4 Argon for harder coding and reasoning, and Gemini 3.5 Flash-Lite for lighter tasks.
Model availability was checked on October 14, 2026 against Google’s model list and the haimaker catalog. Argon is a new paid candidate, not a model we have already benchmarked in OpenClaw.
The quick answer
| Model | Cost tier | Context in haimaker catalog | Best workload to test |
|---|---|---|---|
| Gemini 3.8 Flash | Free directly from Google + paid | About 1M input tokens | Everyday agent work |
| Gemini 4 Argon | Paid | About 1M input tokens | Difficult coding and reasoning |
| Gemini 3.5 Flash-Lite | Free directly from Google + paid | About 1M input tokens | Summaries and lightweight processing |
| Gemini 3.1 Pro | Older preview | Check the active endpoint | Existing integrations |
Gemini 3.8 Flash — the default candidate
Start here for the everyday coding and tool-use comparison. Check correct completion, tool arguments, and retries. Google lists a free tier, but your project’s quota determines how much an agent can run before it needs paid access.
Gemini 4 Argon — the hard-problems candidate
Argon became generally available on October 13, 2026. Try it on difficult tasks that Flash fails, then compare total cost and review time. It is paid-only. Provider and gateway output limits can differ, so use the limits of the endpoint you actually configure.
Gemini 3.1 Pro — an older preview option
Existing users can keep a working integration, subject to provider availability. It should no longer be described as the newest flagship. Include the current Flash and Argon models when you re-evaluate it.
Gemini 3.5 Flash-Lite — lighter work
Test Flash-Lite for classification, file summaries, and other bounded jobs. A cheaper model still needs to follow your tool schemas and instructions; retries can erase the savings.
Legacy models and free-tier access
Google lists Gemini 2.0 Flash as shut down and restricts 2.5-family access to existing users. They should not lead a new setup guide. Use current model IDs, and check Google’s pricing plus your AI Studio quota for free access.
The 1M context advantage
The current candidates above have large input windows. That helps with long specifications and repository reads, but it does not guarantee correct recall or leave enough output space for a large patch. Check the endpoint’s input and output limits separately.
Use only the context your task needs. Carrying unnecessary files through every agent turn increases prompt-processing time and cost. Other current model families also offer large windows; context length alone does not establish which model is best.
Setup in OpenClaw
Running through haimaker.ai
These Gemini candidates are available through haimaker.ai with a single API key. If you’re already using haimaker for other providers, you can access Gemini models without a separate Google account:
{
"models": {
"providers": {
"haimaker": {
"baseUrl": "https://api.haimaker.ai/v1",
"apiKey": "your-haimaker-api-key",
"api": "openai-completions"
}
}
}
}
This gives you Gemini alongside Claude, GPT, DeepSeek, Grok, and dozens of other models through one provider.
And if hand-editing openclaw.json isn’t your thing, the connect CLI handles it in one line: npx -y @haimaker/connect --openclaw writes the provider plus your Gemini allowlist, and --model google/gemini-3.8-flash pins Flash as the default.
Getting any Gemini model running takes about two minutes.
1. Get your Google API key
Go to Google AI Studio and create an API key. Check the available models and quota for your project before relying on free access.
2. Add Google as a provider
Open ~/.openclaw/openclaw.json and add Google to your providers:
{
"models": {
"providers": {
"google": {
"baseUrl": "https://generativelanguage.googleapis.com/v1beta",
"apiKey": "your-google-api-key",
"api": "openai-completions"
}
}
}
}
3. Add models to the allowlist
In the same file, add the models you want:
{
"agents": {
"defaults": {
"models": {
"google/gemini-3.8-flash": {},
"google/gemini-4-argon": {},
"google/gemini-3.5-flash-lite": {}
}
}
}
}
4. Apply the config
Run openclaw gateway config.apply and switch models with /model during a session.
What I’d do
Run one routine task and one difficult failure-recovery task on Gemini 3.8 Flash. Compare the difficult task with Argon and the lighter task with Flash-Lite. Keep the combination that completes your work at an acceptable cost. Recheck the exact IDs and quota when you configure the agent.