xAI Releases Grok 4.3 with Price Cuts and Imagine Agent Mode
Elon Musk's xAI unveiled Grok 4.3, a new AI model targeted at developers for real tasks. This version manages web searches, code execution, and document creation on its own. The company also introduced a beta for Grok Imagine Agent Mode to support creative workflows.
xAI developer Eric Jiang described Grok 4.3 as suited for developers and businesses. It emphasizes speed, low costs, and tool use. The model performs web search, X search, Python code execution, and file search with RAG. It produces Excel files, PDFs, and PowerPoint decks.
Grok 4.3 processes at 100 tokens per second. It features a one-million-token context window. Costs stand at $1.25 per million input tokens and $2.50 per million output tokens. The knowledge cutoff is December 2025. Reasoning comes standard: the model thinks before each response, and those tokens count toward output billing. Access occurs through OpenRouter, the xAI API, and the Hermes agent.
Benchmark Results and Cost Savings
Independent tester Artificial Analysis gave Grok 4.3 a 53 on the Intelligence Index. This exceeds Muse Spark and Claude Sonnet 4.6. It marks four points over Grok 4.20. Still, top models from OpenAI and Anthropic score higher.
Input pricing fell about 40 percent from Grok 4.20. Output pricing dropped about 60 percent. Artificial Analysis places it on the Pareto frontier for best performance-cost balance. A full benchmark run costs $395. That compares to $3,959 for GPT-5.5 and $4,811 for Claude Opus 4.7.
On GDPval-AA, which tests real-world knowledge work, Grok 4.3 reached an Elo score of 1,500. This gain of 321 points puts it past Google's Gemini 3.1. OpenAI's GPT-5.5 leads by 276 Elo points.
Results varied elsewhere. Andon Labs tested AI on a snack vending machine operation. They noted issues with autonomous agent tasks. The model idled at times. Andon Labs said, "It seems to have narcolepsy problems, preferring to sleep for multiple days in a row over taking actions."
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
Val's AI ranked Grok 4.3 first on CaseLaw and first on CorpFin. It placed 13th on general coding benchmarks and harder math problems.
xAI, founded by Elon Musk in 2023 as a rival to OpenAI, builds the Grok series on its own infrastructure. The company aims to advance scientific discovery through AI. Past versions like Grok 4.20 set benchmarks in speed and cost efficiency.
Grok Imagine Agent Mode Launch
xAI rolled out Agent Mode for Grok Imagine in beta via the Grok web interface. Users activate it in the input field at the bottom left. A paid account is necessary.
Unlike single prompts, this mode tackles full creative projects. An AI agent plans, generates, edits, and revises in an open workspace. xAI suggests uses such as a one-minute movie, a manga set, or product stories.
This addition builds on Grok Imagine's image generation. It enables cohesive production for longer efforts. The web interface makes it straightforward for paid users to start.
Grok 4.3 arrives amid competition in AI models. xAI focuses on practical tools for work and creation. Lower prices position it for wider developer adoption.