xAI has released Grok 4.3, a faster, cheaper, and more capable model designed for practical developer and business use. The update also introduces an agent-based image generation tool called Grok Imagine Agent Mode.
Developer Eric Jiang announced that Grok 4.3 emphasizes speed, low cost, and tool usage. The model can independently perform web searches, X searches, Python code execution, and file-based retrieval (RAG), and it can generate Excel files, PDFs, and PowerPoint presentations.
Grok 4.3 delivers 100 tokens per second and supports a one-million-token context window. Pricing is set at $1.25 per million input tokens and $2.50 per million output tokens, with a knowledge cutoff of December 2025. Reasoning is enabled by default—the model “thinks” before responding, and reasoning tokens are billed at the same rate as standard output tokens. The model is available via OpenRouter, the xAI API, and the Hermes agent.
Lower Prices, Mid-Tier Benchmark Scores
According to independent benchmarking service Artificial Analysis, Grok 4.3 achieved a score of 53 on the Intelligence Index. This slightly surpasses Muse Spark and Claude Sonnet 4.6, and is four points higher than its predecessor Grok 4.20, though still far behind flagship models from OpenAI and Anthropic.
Input costs have dropped roughly 40% and output costs about 60% compared to Grok 4.20, placing Grok 4.3 on the Pareto frontier for cost-performance balance. A full benchmark run costs $395, versus $3,959 for GPT-5.5 and $4,811 for Claude Opus 4.7.
On the GDPval-AA benchmark, which measures AI performance on real-world knowledge work, Grok 4.3’s Elo score jumped 321 points to 1,500, surpassing Google’s Gemini 3.1, though OpenAI’s GPT-5.5 still leads by 276 Elo points.
However, other benchmarks revealed weaknesses. Andon Labs reported setbacks on autonomous agent tasks, with the model sometimes remaining idle instead of taking action, humorously noting it “seems to have narcolepsy problems.” Val’s AI ranked Grok 4.3 first on CaseLaw and CorpFin but found it struggles with general coding and harder math problems, landing in 13th place.
New Agent Mode for Grok Imagine
xAI also launched Agent Mode for Grok Imagine, now in beta on the Grok web interface. Unlike single-prompt generation, this mode handles extended creative projects by planning, generating, editing, and revising content in an open workspace. xAI suggests it works well for creating a one-minute movie, a manga set, or product stories.
Agent Mode can be accessed via the Grok web interface and enabled in the input field at the bottom left. A paid account is required.