xAI

The Grok family

xAI's Grok models, which launched touting real-time knowledge from the X platform.

Only versions we have a record for. Missing ones appear as records are added.

11versions2023–2026Latest · Grok 4.6

Grok arrived in November 2023 as xAI's first product. From the start the company has published benchmark scores it measured itself, and the records here mark them as self-reported.

2024 brought Grok-1.5, which reads much longer input at once, and Grok-1.5V, the first model in the line that understands images. In 2025 came Grok 3, with a reasoning mode that thinks before answering, and Grok 4, which scaled up the compute spent on reinforcement learning.

From late 2025 the line split by use: a coding-only model, and a model that folds reasoning and ordinary answers into one set of weights. In 2026 xAI said Grok 4.5 was trained together with Cursor, the AI coding-tool company, and Grok 4.6 focused on long-running agent tasks.

Made by AIThe family notes were written by AI (Claude); each version links to its record, where the source is. Published after automated checks; the owner doesn't read every item.How it's made

At a glance

Where labels would overlap, only the dot is drawn — point at it to see the name; every version is in the list below.

11 versions

In release order. Each date comes from the record that documents it — open a name to see the record.

  1. Grok

    xAI's first product; the main model was Grok-1

    Text
  2. Grok-1.5

    Context expanded to 128,000 tokens

    Text
  3. Grok-1.5V

    Preview of xAI's first multimodal model, with RealWorldQA

    Multimodal
  4. Grok-2 / Grok-2 mini

    Beta release; image generation via an outside model

    Text
  5. Grok 3 / Grok 3 mini

    Think-button reasoning mode and first agent, DeepSearch

    Reasoning
  6. Grok 4 / Grok 4 Heavy

    Scaled up RL compute toward pretraining levels

    Reasoning
  7. grok-code-fast-1

    Low-cost model built for agentic coding

    Code
  8. Grok 4 Fast

    Reasoning and non-reasoning modes in one set of weights

    Reasoning
  9. Grok 4.1

    Update that targeted emotional and creative chat

    Text
  10. Grok 4.5

    Trained jointly with the coding company Cursor

    Reasoning
  11. Grok 4.6

    Update focused on long-running agents

    LatestReasoning