MiniMax is the cheapest way to run a capable model in OpenClaw right now. As of April 2026, there are five MiniMax models worth considering, and most people are trying to figure out which one is worth the setup effort.
Short version: start with M2.5, swap to Lightning if speed matters, ignore the rest.
The quick answer
| Model | Input/Output Cost | Context | Best For |
|---|---|---|---|
| MiniMax M2.5 | $0.12 / $1.00 | 196K | Default — cheap, capable |
| MiniMax M2.5 Lightning | $0.30 / $2.40 | 1M | Speed + long context |
| MiniMax M2.1 | $0.27 / $0.95 | 196K | Budget alternative |
| MiniMax M2.1 Lightning | $0.30 / $2.40 | 1M | Legacy — use 2.5 Lightning |
| MiniMax M2 | $0.30 / $1.20 | 200K | Legacy — use 2.5 |
Start with M2.5. Reach for M2.5 Lightning when you need speed or a 1M context window. Skip the older versions.
MiniMax M2.5 — the default pick
At $0.12/M input and $1/M output, M2.5 is in a different cost tier than almost every other model that works well with OpenClaw. For comparison, Claude Sonnet 4.6 costs $3/$15. That’s about 25x more on input and 15x more on output. MiniMax M2.5 won’t match Sonnet on the hardest reasoning tasks, but it gets surprisingly close on day-to-day coding.
The 196K context window is enough for most real work. Function calling is reliable. Reasoning is supported and actually helps on multi-step tasks. The model is particularly good at staying coherent across long conversations, which matters when you’re using OpenClaw interactively over an hour.
Where it falls short: novel algorithms, hard debugging where the model needs to form and test hypotheses, and tasks that require the absolute latest world knowledge. For everything else (refactoring, file edits, documentation, code explanation, tool-heavy workflows), it’s the cheapest way to get real work done.
I’ve been routing 70–80% of my OpenClaw traffic to M2.5 and only escalating to Sonnet 4.6 or GPT-5.4 when the cheaper model gets stuck. The monthly savings are significant.
MiniMax M2.5 Lightning — when speed matters
$0.30/M input, $2.40/M output, 1M token context. Lightning is the faster variant: higher throughput, lower latency, and 5x the context window of regular M2.5.
You pay for it. Output tokens cost 2.4x more than regular M2.5, which adds up on code-generation tasks. But if you’re loading huge contexts (whole monorepos, long documents, giant log files) or running interactive workflows where latency is visible, the upgrade is worth it.
The trade-off vs. regular M2.5 is mostly economic: Lightning costs more but handles the 800K–1M token range that M2.5 can’t. If your inputs fit in 196K, stay on M2.5. If they don’t, Lightning is the call.
MiniMax M2.1 — the budget backup
$0.27/M input, $0.95/M output, 196K context. M2.1 predates the 2.5 release and is slightly cheaper on output tokens but a bit behind on quality. Not by a lot, but enough that I’d rather pay the few extra cents for M2.5.
The one case for M2.1 is high-volume batch work where output tokens dominate the bill and quality is “good enough.” If you’re summarizing thousands of documents or generating commit messages at scale, M2.1 can save real money. For coding with OpenClaw, go with M2.5.
Legacy models to skip
- MiniMax M2 ($0.30/$1.20): Replaced by M2.5 at a lower price with better quality. No reason to use it.
- MiniMax M2.1 Lightning ($0.30/$2.40): Same price as M2.5 Lightning but older. Always prefer the newer version.
Setup in OpenClaw
The fastest path to MiniMax in OpenClaw is through haimaker.ai. MiniMax models are available alongside Claude, GPT, Gemini, Grok, and open-source models through a single API key.
1. Get your haimaker.ai API key
Sign up at haimaker.ai and copy your key from the dashboard. New accounts come with credits to test things out.
2. Add haimaker as a provider
Open ~/.openclaw/openclaw.json and add:
{
"models": {
"providers": {
"haimaker": {
"baseUrl": "https://api.haimaker.ai/v1",
"apiKey": "your-haimaker-api-key",
"api": "openai-completions"
}
}
}
}
3. Add MiniMax models to the allowlist
{
"agents": {
"defaults": {
"models": {
"haimaker/minimax/minimax-m2.5": {},
"haimaker/minimax/minimax-m2.5-lightning": {},
"haimaker/minimax/minimax-m2.1": {}
}
}
}
}
4. Apply the config
Run openclaw gateway config.apply and switch models with /model during a session.
What I’d do
Set M2.5 as your default. It’s so cheap it’s almost free, and for most coding tasks it’s good enough that you won’t reach for a flagship model unless you really need one. Swap to M2.5 Lightning when your inputs blow past 196K or when latency is holding you back. Leave the older MiniMax versions alone. They’re not worth the setup complexity.
If you’re spending more than $30/month on Claude or GPT, try routing 80% of your OpenClaw traffic to M2.5 for a week. Most people see their bill drop by 60–90% with no noticeable quality regression.