Current as of March 2026. Grok 3 is xAI’s current flagship text model — $3/M input, $15/M output, 131K context on both ends. That 131K output limit is the headline feature; GPT-4o caps at 4K output, which becomes genuinely painful when you’re generating substantial amounts of code or documentation.
Specs
| Provider | xAI |
| Input cost | $3.00 / M tokens |
| Output cost | $15 / M tokens |
| Context window | 131K tokens |
| Max output | 131K tokens |
| Parameters | N/A |
| Features | function_calling, web_search |
What it’s good at
Output capacity
131K max output tokens is unusual at this tier. If you’ve ever hit a model mid-function because GPT-4o’s 4K limit ran out, Grok 3 solves that problem. Entire module rewrites, long reports, multi-file diffs — it can actually finish them.
Speed
Token generation is faster than Claude 3.5 Sonnet. For interactive agent workflows where latency matters, that difference is noticeable.
Where it falls short
Persona bleed
Grok has a baked-in personality that’s harder to suppress than you’d like. System prompts telling it to be neutral and professional help, but informal or sarcastic phrasing still bleeds through in ways that would be unacceptable in customer-facing contexts.
API reliability
The xAI infrastructure is not as stable as Azure or AWS. Rate limit errors and occasional downtime during high-traffic periods are real issues. If you run this in production, build retry logic.
Best use cases with OpenClaw
- Large-scale code generation — The 131K output window means you can prompt for a substantial multi-file structure without the model cutting off mid-function. That alone justifies trying it for the right workloads.
- Input-heavy agent loops — At $3/M input tokens, it’s cheaper than GPT-4o for agents that repeatedly ingest large context payloads.
Not ideal for
- Strict corporate chatbots — The personality issue is a liability for customer-facing roles. You can prompt around it but you can’t fully eliminate it.
- Zero-failure production systems — Until xAI’s API stability catches up to the established providers, it’s a secondary choice for anything mission-critical.
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
Use the OpenAI-compatible provider in OpenClaw and point the base URL to https://api.x.ai/v1. You will need to manually set the model ID to xai/grok-3 in your configuration file.
{
"models": {
"mode": "merge",
"providers": {
"xai": {
"baseUrl": "https://api.x.ai/v1",
"apiKey": "YOUR-XAI-API-KEY",
"api": "openai-completions",
"models": [
{
"id": "grok-3",
"name": "Grok 3",
"cost": {
"input": 3,
"output": 15
},
"contextWindow": 131072,
"maxTokens": 131072
}
]
}
}
}
}
How it compares
- vs Claude 3.5 Sonnet — Sonnet has superior instruction following and a more professional tone, but Grok 3 wins on output length and raw speed.
- vs GPT-4o — GPT-4o offers better tool-calling reliability and a more stable API, while Grok 3 is cheaper for input-heavy workloads at $3 per million tokens.
Bottom line
If you need to generate long outputs and can stomach some API instability, Grok 3 is worth it. The 131K output limit is a real differentiator. Just don’t use it anywhere the persona bleed would be a problem.
For setup instructions, see our API key guide. For all available models, see the complete models guide.