xAI launches Grok 4.3 with always-on reasoning and voice cloning suite

xAI has released Grok 4.3, a new base large language model featuring what the company calls "always-on reasoning," alongside a voice cloning suite called Custom Voices. Unlike earlier versions where reasoning could be toggled, Grok 4.3 is built to think before responding on every query. The model supports a 1 million-token context window and accepts both text and image inputs, outputting text. It is optimized for agentic workflows and can generate complex files such as spreadsheets, PDFs, and PowerPoint decks. Early demonstrations included a multi-sheet OSRS Sailing Combat DPS analyzer built in roughly six minutes and a 12-page SpaceX product report with branding and tables. API pricing is set at $1.25 per million input tokens and $2.50 per million output tokens, with costs doubling for inputs exceeding 200,000 tokens. The model also introduces reasoning tokens billed at standard completion rates, prompt caching at $0.20 per million tokens, and tool invocation fees including $5.00 per 1,000 calls for web search or code execution. A $0.05 fee applies to requests blocked by safety filters. Independent evaluator Artificial Analysis reported that Grok 4.3 is roughly 40% cheaper on input and 60% cheaper on output than its direct predecessor, Grok 4.2. Vals AI placed it first on the CaseLaw v2 and CorpFin benchmarks, though other tests showed regression on general coding tasks and agentic consistency. One benchmark operator described the model as having "narcolepsy problems" in a simulation environment. xAI also launched a voice cloning API that clones voices from 120-second audio clips. The feature is restricted to the United States, excluding Illinois, and is available through a web console and Enterprise API. Text-to-Speech is priced at $4.20 per 1 million characters, and the Voice Agent API at $3.00 per hour. The model's training knowledge extends to December 2025, and it includes live web and X search tools to access current information.