English
English
Grok is a large language model launched by xAI, which includes Grok-0, Grok-1, Grok-1.5, Grok-1.5V, Grok-2, Grok-2 mini, Grok-3, Grok-3 mini and other models.
| Model | Release time | Features |
|---|---|---|
| Grok-0 | August 18, 2023 | xAI's first flagship model, 33 billion parameters |
| Grok-1 | November 3, 2023 | 314 billion parameters, using technologies such as quantitative weights, hybrid expert architecture and rotational position embedding, the context size is 8192 tokens, the vocabulary size is 131072, and the score is 73% in the MMLU benchmark test |
| Grok-1.5 | March 28, 2024 | Adding long text capabilities on Grok-1 |
| Grok-1.5V | April 12, 2024 | Add visual ability based on Grok-1, and reasoning can be done based on interlaced natural language and images |
| Grok-2 | August 13, 2024 | Stronger inference ability and upgraded performance. Use Black Forest Labs' FLUX.1 for image generation |
| Grok-2 mini | August 13, 2024 | Grok-2's smaller version, balancing speed and answer quality |
| Grok-3 | February 18, 2025 | The flagship model, designed for advanced reasoning and professional tasks, introduces "thinking chain" technology, supports multimodal input, has a self-correction mechanism, and performs excellently in mathematical, scientific and programming benchmarks, surpassing models such as GPT-4o, Gemini and DeepSeek. Equipped with DeepSearch, which allows you to retrieve information from multiple sources and compile concise reports |
| Grok-3 mini | February 18, 2025 | Cost-effective variant designed for STEM tasks requiring less world knowledge, with better resource consumption and speed, performing in AIME 2024 and LiveCodeBench with 95.8% accuracy and 80.4% respectively |