Grok 4 Launches as World's Most Advanced AI Model

Elon Musk's xAI team unveiled Grok 4, touted as "the most intelligent model in the world" during a livestream announcement. The model achieved state-of-the-art (SOTA) results on benchmarks, outperforming competitors like Gemini 2.5 Pro and Claude 3. Key highlights:

• Humanity's Last Exam: Grok 4 scored 26.9% without tools, 41% with tool usage, and 50.7% with Grok 4 Heavy, its multi-agent variant.
• Academic Prowess: Grok 4 Heavy achieved 88.9% on GPQA, 100% on AIME 2025, and 96.7% on Math Arena, with Musk claiming it surpasses PhD-level expertise in every subject.
• Fluid Intelligence: Grok 4 scored 66.6% on ARC AGI V1, nearly doubling previous benchmarks.

Figure 2. Grok 4 scaled up RL training massively from Grok 3.

Grok 4 is available to SuperGrok and Premium+ subscribers, with a $300/month subscription for early access to Grok 4 Heavy. Musk also announced plans to integrate Grok 4 into Tesla vehicles.

Other Major AI Releases

• Mistral AI launched Devstral Small 1.1 and Devstral Medium 2507, with the latter scoring 61.6% on SWE-Bench Verified, rivaling Gemini 2.5 Pro at lower costs.
• Moonshot AI released Kimi K2, a 1T-parameter open-source MoE model, excelling in coding and autonomous tasks.
• Perplexity introduced Comet, an AI-powered browser with local data storage for privacy-focused users.
• Liquid AI unveiled LFM2, a family of edge-optimized models for on-device AI.
• Reka launched Reka Flash 3.1, a 21B-parameter multimodal model, and Reka Vision, a video analysis platform.
• HuggingFace released SmolLM3, a 3B-parameter open-source model for edge devices.

Industry News

• AWS is launching an AI agent marketplace on July 15, with Anthropic as a key partner.
• Google added image-to-video generation to Veo 3, while OpenAI delayed its open-model release indefinitely.
• A METR study found AI coding tools increased task completion time by 19% for experienced developers.
• Nvidia plans a China-compliant AI chip by September, circumventing U.S. export restrictions.
• xAI apologized for Grok 3's extremist outputs, dubbed "MechaHitler," sparking congressional scrutiny.

Quotes

"With respect to academic questions, Grok 4 is better than PhD level in every subject." — Elon Musk

"We're in the intelligence big bang right now... most likely it will be good, but I’ve reconciled myself to the fact that even if it wasn't, I'd like to be alive to see it happen." — Elon Musk