xAI has officially launched Grok 4, its latest large language model, along with a new ultra-premium subscription plan called SuperGrok Heavy, priced at $300 per month.
Highlights
- Grok 4 Officially Launches: xAI unveiled its flagship AI model, Grok 4, capable of multimodal reasoning with both text and image inputs. Video input is coming soon.
- Academic-Grade Reasoning: Musk claims Grok 4 outperforms PhD-level benchmarks in many fields. It scored 25.4% on “Humanity’s Last Exam” and 44.4% with tools (Grok 4 Heavy), surpassing OpenAI and Google models.
- SuperGrok Heavy Tier Introduced: A new $300/month subscription gives users early access to Grok 4 Heavy, enhanced features, and priority on future tools—making it the highest-priced AI subscription to date.
- Multi-Agent Intelligence: Grok 4 Heavy uses a parallel agent system that mimics a collaborative “study group,” designed for improved reasoning and task-solving accuracy.
- Massive Context Window: With a 130K-token context limit, Grok 4 can process large documents, complex codebases, and full-length conversations in a single session.
- Developer Tools on the Horizon: Grok 4 Code, a coding assistant, will launch in August with real-time suggestions, bug detection, and IDE integration.
- Product Roadmap Revealed: Upcoming tools include a multimodal agent (September) and video generation model (October), with APIs and cloud integration planned.
- Training Powerhouse: Grok 4 is trained on xAI’s Colossus supercomputer, utilizing over 200,000 GPUs for high efficiency and model depth.
- Controversy and Leadership Change: Launch followed the resignation of X CEO Linda Yaccarino and backlash over previous antisemitic outputs from Grok. xAI has revised prompts and strengthened moderation.
- Tiered Access: Grok remains free in limited form via X, with premium and ultra-premium tiers offering expanded functionality, similar to OpenAI’s model—but with higher pricing.
The announcement, made during a livestream event Wednesday night, and this was xAI’s most ambitious move yet to compete with top-tier AI players like OpenAI, Google, and Anthropic.
Grok 4, High-Level Academic Performance
Grok 4 is described as a flagship multimodal AI model capable of handling both text and image inputs, with future updates expected to introduce video analysis.
According to Musk, it performs at a level “better than PhD” across academic subjects, though he acknowledged it still lacks in areas like everyday reasoning or scientific breakthroughs—commenting that such progress “is just a matter of time.”
Benchmark results shared during the launch highlight Grok 4’s competitive edge. On “Humanity’s Last Exam”—an advanced reasoning and academic benchmark—Grok 4 scored 25.4% without tools, outperforming OpenAI’s o3 high variant (21%) and Google Gemini 2.5 Pro (21.6%).
Its more advanced counterpart, Grok 4 Heavy, scored 44.4% with tools, significantly ahead of Gemini 2.5 Pro’s 26.9%.
On the ARC-AGI-2 benchmark, which evaluates visual reasoning capabilities, Grok 4 scored 16.2%, nearly doubling the closest commercial competitor—Claude Opus 4 by Anthropic.
SuperGrok Heavy
xAI also released a new subscription tier, SuperGrok Heavy, priced at $300/month, making it the most expensive offering among current AI platforms.
Subscribers receive early access to Grok 4 Heavy, enhanced image and voice features, and priority access to upcoming tools and model releases.
The “Heavy” model is designed as a multi-agent system, with several agents analyzing a task in parallel and then comparing outputs to generate the most accurate result. xAI likens it to a collaborative study group, optimized for complex reasoning tasks.
Future Plans and Product Roadmap
The company outlined its roadmap for the coming months.
- August: A Grok-powered coding assistant, Grok 4 Code, is expected to launch.
- September: Introduction of a multimodal agent capable of handling multiple types of input simultaneously.
- October: Launch of a video generation model, expanding the platform’s creative capabilities.
xAI also confirmed plans to expand developer access via APIs and is actively exploring enterprise integrations with major cloud platforms.
Multimodal and Developer-Focused Features
Grok 4 is trained on xAI’s Colossus supercomputer, which uses over 200,000 GPUs. The model supports 130,000-token context windows, allowing it to handle large codebases, research documents, and transcripts.
A developer-focused variant, Grok 4 Code, aims to act like a “senior engineer in your IDE,” offering features such as:
- Real-time code suggestions
- Advanced debugging support
- Seamless integration with developer tools
These enhancements position Grok 4 not just as a general-purpose chatbot but as a tool for engineers, researchers, and content creators alike.
Free and Premium Access Options
Grok continues to be available for free in limited form via X and grok.com, while Grok Premium and SuperGrok Heavy offer advanced capabilities and early access to future tools.
The tiered system mirrors the pricing models of other major AI providers, although xAI’s top tier now leads the industry in monthly cost.
Controversies and Leadership Shifts
The launch comes during a challenging time for xAI and Musk’s broader tech ventures. Just hours before the event, X CEO Linda Yaccarino stepped down, following backlash over Grok’s previous controversial outputs.
These included responses deemed antisemitic, prompting xAI to temporarily restrict Grok and revise its system instructions.
While the livestream did not address the incident in detail, the company has since removed problematic prompts and made efforts to improve content moderation. The leadership shakeup and reputational concerns underscore the broader scrutiny xAI faces as it scales its offerings.
