Current as of March 2026. Grok 3 Mini sits in an interesting spot: $0.30/M input, $0.50/M output, 131K context window, and it has a reasoning mode. That combination is unusual at this price. Claude 3.5 Haiku costs $0.80/$4.00 for roughly similar capability — Mini is dramatically cheaper.
Specs
| Provider | xAI |
| Input cost | $0.30 / M tokens |
| Output cost | $0.50 / M tokens |
| Context window | 131K tokens |
| Max output | 131K tokens |
| Parameters | N/A |
| Features | function_calling, reasoning, web_search |
What it’s good at
Output capacity
Most small models cap output at 4K–8K tokens. Grok 3 Mini goes to 131K. For generating large code files or long technical documents, that matters. You stop breaking tasks into chunks.
Reasoning at low cost
The reasoning capability here is real, not just marketing. It handles logical branching in OpenClaw agent flows noticeably better than GPT-4o-mini. You’re getting something that previously required a much more expensive model.
Price
$0.50/M output is extremely cheap. For high-frequency agent polling where the model is constantly responding to tool results, the cost difference between this and Haiku is substantial over time.
Where it falls short
API stability
Same infrastructure caveat as all xAI models: latency spikes happen, and you need retry logic. This is more of an issue in OpenClaw than in one-off scripts because agents can get stuck waiting.
Refusal behavior
The safety filters are inconsistent. It occasionally refuses tasks in automated loops that a human would find completely benign. This can silently break an agent workflow if you’re not monitoring for refusal responses.
Best use cases with OpenClaw
- Bulk log analysis — The 131K context window lets you dump large log files into a single prompt for error pattern detection. The price makes it practical to do this at scale.
- Sub-agent logic nodes — Intermediate nodes in OpenClaw that need tool-calling and basic reasoning but don’t need flagship-level capability. This is the right model for the job when $15/M is overkill.
Not ideal for
- Safety-sensitive automation — The inconsistent guardrails mean you can’t fully trust the refusal behavior in both directions. It might refuse something harmless or let something through.
- Creative writing — The prose is dry and repetitive. Claude 3.5 Haiku is meaningfully better here even at higher cost.
Run it through Haimaker
Skip juggling API keys. One Haimaker key gives you access to every model on the platform. Tell OpenClaw:
Add Haimaker as a custom provider to my OpenClaw config. Use these details:
- Provider name: haimaker
- Base URL: https://api.haimaker.ai/v1
- API key: [PASTE YOUR HAIMAKER API KEY HERE]
- API type: openai-completions
Add the auto-router model:
- haimaker/auto (reasoning: false, context: 128000, max tokens: 32000)
Create an alias "auto" for easy switching. Apply the config when done.
Or skip model selection entirely — Haimaker’s auto-router picks the best model for each task so you don’t have to.
OpenClaw setup
Set your provider to OpenAI-compatible and use the base URL https://api.x.ai/v1. Ensure your API key is correctly mapped to the xai/grok-3-mini model ID in your environment variables.
{
"models": {
"mode": "merge",
"providers": {
"xai": {
"baseUrl": "https://api.x.ai/v1",
"apiKey": "YOUR-XAI-API-KEY",
"api": "openai-completions",
"models": [
{
"id": "grok-3-mini",
"name": "Grok 3 Mini",
"cost": {
"input": 0.3,
"output": 0.5
},
"contextWindow": 131072,
"maxTokens": 131072
}
]
}
}
}
}
How it compares
- vs GPT-4o-mini — Grok 3 Mini is more expensive on input ($0.30 vs $0.15) but provides superior reasoning and a much larger output capacity.
- vs Claude 3.5 Haiku — Grok 3 Mini destroys Haiku on price, costing $0.30/$0.50 compared to Haiku’s $0.80/$4.00, while offering similar logic performance.
Bottom line
If you’re running lots of agent nodes that need some reasoning ability, Grok 3 Mini is the cheapest way to get it. The xAI reliability caveats apply, but the cost-to-capability ratio is hard to match.
For setup instructions, see our API key guide. For all available models, see the complete models guide.