Grok 4
2025-07-09 · Publisher : xAI · Domain : Artificial Intelligence
xAI launches Grok 4, presented as its most intelligent model, with native tool use, real-time search and a 256,000-token context window in the API.
What changes
- Reinforcement learning scaled up on Colossus (a 200,000-GPU cluster), with a claimed 6x gain in compute efficiency.
- Native tool use: code interpreter, web browsing, advanced search on X (keywords, semantic, media).
- Improved voice mode with camera vision.
- Multimodal API (text and image) with live search on X, the web and the news.
### Specifications - Context: 256,000 tokens (API). - Modalities: text and vision. - Availability: SuperGrok and Premium+ subscribers, xAI API; hyperscalers announced “soon”. Certifications cited: SOC 2 Type 2, GDPR, CCPA. - API identifier: grok-4 (grok-4-0709 cited by xAI in a later post). - Knowledge cutoff: not disclosed. ### Pricing Launch pricing recorded from an aggregator (the post does not publish it): $3.00 input, $15.00 output, $0.75 cached input, per million tokens. Long-context tier: not verified. ### Benchmarks Scores published by the vendor (x.ai post). | Benchmark | Score | |---|---| | ARC-AGI v2 | 15.9% | | Vending-Bench (average net worth over 5 runs) | $4,694.15 | | Humanity's Last Exam (Heavy, text subset) | 50.7% | | USAMO 2025 (Heavy) | 61.9% | ### Key takeaways Grok 4 marks xAI's shift to large-scale reinforcement reasoning and native tool use, with a 256,000-token API.
Links
Tags : LLM, xAI, Grok, Grok 4, Raisonnement