xAI's flagship. Top math/science benchmarks, live X data, "unhinged" mode that actually answers.
Grok 3 is the third-generation model from xAI, Elon Musk's AI lab, released in 2025 and iterated through 2026. It was trained on the largest GPU cluster ever built (200,000 H100s) and currently sits at or near the top of the math and science benchmarks. It's also the only frontier model with native access to live X (Twitter) data — useful for trending topics, sentiment, and breaking news.
The model's signature feature is its "unhinged" mode, which actually answers questions other models refuse. Combined with live X data and image generation (Aurora), it's positioned as the "AI with fewer guardrails" alternative to GPT-5 and Claude. Whether that's a feature or a bug depends on your use case.
Who it's for: Power users who want live data from X (Twitter), math/science researchers, and anyone who likes a less-filtered AI.
On AIME 2025 and GPQA Diamond, Grok 3 leads the frontier. It also has a dedicated "Think" mode that uses more compute for harder problems.
The only frontier model with native access to the X firehose. Great for trend analysis, sentiment, and breaking news synthesis.
xAI's image model is fast and the photorealism is competitive with Midjourney. No separate subscription needed.
Optional mode that actually answers edgy questions other models refuse. Useful for fiction writing, satire, and brainstorming — but not a default for serious work.
If you need top-tier math/science reasoning, live X data, or a model that won't refuse your creative writing prompts, Grok 3 is worth the $30/mo. For most general-purpose writing and analysis, Claude 4 Opus or GPT-5 is a better fit.