xAI Launches Grok 4.3: Affordable AI, But Behind OpenAI

xAI, Elon Musk's company, recently unveiled Grok 4.3, a developer-focused artificial intelligence model. This model is designed to autonomously perform various tasks such as web research, code execution, and document creation. Grok 4.3 outperforms its predecessor in real-world performance tests, but it still lags behind the leading models from OpenAI and Anthropic.

In addition to its enhanced capabilities, Grok 4.3 is offered at a reduced cost, making it more accessible for developers and businesses. The model can handle tasks such as web research, executing Python code, and generating Excel, PDF, and PowerPoint files. It operates at a speed of 100 tokens per second and has a context window of one million tokens. Pricing is set at $1.25 per million input tokens and $2.50 per million output tokens, with a knowledge cutoff date set for December 2025. Reasoning is integrated by default, meaning Grok 4.3 "thinks" before responding to each request, and reasoning tokens are charged at the same rate as standard output tokens. The model is available via OpenRouter, xAI's API, and the Hermes agent.

Performance and Costs of Grok 4.3

According to Artificial Analysis, an independent benchmarking service, Grok 4.3 achieved a score of 53 on the Intelligence Index. This score places it slightly above Muse Spark and Claude Sonnet 4.6, but well behind the flagship models from OpenAI and Anthropic. Input costs have decreased by about 40% and output costs by approximately 60% compared to the previous version, Grok 4.20. A full test of Grok 4.3 costs $395, compared to $3,959 for GPT-5.5 and $4,811 for Claude Opus 4.7.

On the GDPval-AA benchmark, which evaluates AI performance on practical tasks, Grok 4.3 achieved an Elo score of 1,500, surpassing Google Gemini 3.1. However, it remains behind GPT-5.5, which maintains a lead of 276 Elo points. Other benchmarks, such as those from Andon Labs, reveal weaknesses in autonomous agent tasks, where Grok 4.3 may remain inactive. "It seems to have issues with narcolepsy, preferring to sleep for several days in a row rather than act," reports Andon Labs. Val's AI ranks Grok 4.3 first on CaseLaw and first on CorpFin, but notes difficulties on general coding benchmarks and more complex math problems, where it ranks 13th.

The Grok Imagine Agent Mode

In parallel, xAI has introduced the Agent Mode for Grok Imagine, currently in beta. This mode is designed to manage longer creative projects, allowing an AI agent to plan, generate, edit, and revise content in an open workspace. This mode is particularly useful for creating short films, manga, or product stories. Accessible via the Grok web interface, activating the Agent Mode requires a paid account. This feature aims to provide consistent creative production workflows, enabling users to work on complex projects with continuous AI assistance.