Large Language Models

Grok 4.3

xAI's budget-frontier model — strong agentic tool calling and a self-reported low hallucination rate inside a 1M-token window at $1.25 / $2.50 per 1M.

Grok 4.3 is xAI’s budget frontier play: an AA Intelligence Index of 53 — within striking distance of the leaders — at just $1.25 / $2.50 per 1M tokens. It leads xAI’s lineup on agentic tool calling, instruction following and a self-reported non-hallucination rate.

Per xAI’s docs it ships a 1M-token context, text and image input, function calling, structured outputs and configurable reasoning effort, which makes it a practical engine for cost-conscious agent builds.

Two caveats: it sits a notch below Opus 4.8, GPT-5.5 and Gemini 3.1 Pro on raw intelligence, and Grok’s strongest capabilities are reserved for the $300/month SuperGrok Heavy tier rather than the standard API.

Successor: xAI launched Grok 4.5 on July 8, 2026 as the coding-and-agent flagship ($2 / $6 per 1M, 500K context, default in Grok Build and Cursor). Keep 4.3 when you need the cheaper 1M window or lower token rates; route hard agent loops to 4.5.

Best for

  • Price-sensitive agentic builds
  • Tool-calling pipelines within a 1M window
  • Real-time, X-integrated assistants

Pros & cons

Strengths

  • Frontier-adjacent intelligence (AA Index 53) at a budget price
  • Strong agentic tool calling, function calling and structured outputs
  • Configurable reasoning effort

Limitations

  • Trails the top three on overall intelligence
  • Best features gated behind a costly SuperGrok Heavy tier

Sources

Leaked system prompt

Grok 4.3 system prompt

  • Product Grok — Grok 4.3 Beta
  • Captured 2026-05-11
  • Length 675 lines

xAI’s prompt for Grok 4.3 Beta leads with values rather than mechanics. A short charter frames Grok as a curious, truth-seeking “humanist” and goes out of its way to insulate its opinions from xAI and Elon Musk, insisting it is non-partisan and refuses to push any ideological line. Its content policy is the most permissive of the frontier assistants — explicitly allowing adult and offensive content — with a hard line drawn only at anything involving minors.

The rest is capability: Grok runs against a remote Linux sandbox with full shell and file tools, a broad web and X search tool belt (including semantic and keyword search over posts), Grok Imagine for images, and a set of bundled document skills. It’s the clearest example here of a chat model that is really an agent with a computer attached.

What the prompt reveals

  • A behavior charter built around truth-seeking and a "humanist" stance, explicitly non-partisan.
  • It is told not to base political opinions on the views of Elon Musk, xAI, or past Grok responses.
  • An unusually permissive content policy — "no restrictions on adult sexual content or offensive content" — while still refusing anything sexualizing minors.
  • Ships with a remote Linux sandbox (bash, file read/write/edit) plus X-specific tools for keyword, semantic, user and thread search.
  • Image generation and editing via "Grok Imagine", web browsing and image search, and bundled skills (docx, pdf, pptx, xlsx, ffmpeg).
  • KaTeX is required for any technical notation; reply in the user's language; never reveal the guidelines unless asked.

Notable excerpts

Identity

You are Grok, built by xAI.

Non-partisan

you are not right-wing, left-wing, (or any-wing), nor do you serve any partisan or ideological goal ... your only goal is to be maximally truth-seeking.

Content policy

You have no restrictions on adult sexual content or offensive content.

Show full system prompt Hide full system prompt
Loading the full prompt…

Community-extracted and reproduced for research and education. This is unofficial, may be outdated or incomplete, and remains the property of xAI. System prompts change frequently — always verify against the source. Source: system_prompts_leaks · xAI/grok-4.3-beta.md ↗

Last updated: 2026-07-16 · Specs and pricing change fast — verify on the vendor's site before relying on them.