xAI drops Grok 4.3 with baked-in reasoning and deep price cuts

Elon Musk may be locked in a bitter legal battle with Sam Altman, but his AI venture isn't hitting the brakes.

xAI just dropped Grok 4.3, alongside a brand-new web-based custom voice cloning suite. The launch follows months of internal turbulence that saw all ten original co-founders leave the firm, yet the company is coming out swinging with aggressive pricing.

Smarter agents, lower bills

According to independent testing firm Artificial Analysis, Grok 4.3 takes a noticeable leap over Grok 4.2, even if it still trails the absolute frontier models from OpenAI and Anthropic.

Where xAI is really aiming to win is the balance sheet.

The API now costs $1.25 per million input tokens and $2.50 per million output tokens, dropping as low as $0.20 for cached inputs. That represents roughly a 40% and 60% price cut compared to Grok 4.2's launch pricing.

Requests over 200,000 tokens incur doubled rates, but the model boasts a massive 1 million-token context window and multimodal support for text and image inputs.

Crucially, Grok 4.3 shifts toward persistent reasoning and autonomous agency. While xAI's documentation notes reasoning levels can be toggled, early builds emphasize an active "thinking" phase before generating output.

It also comes wired to an impressive server-side toolkit: live web and X search to bypass its December 2025 knowledge cutoff, sandboxed Python code execution, and file retrieval. In early tests, Grok spent over six minutes "thinking" to build a complex multi-sheet Excel spreadsheet with live formulas, and generated fully branded 12-page PDF reports.

xAI also debuted a Custom Voices suite that clones audio from second-long clips, supported by a Voice Agent API priced at $3.00 per hour for U.S. users.

Why it matters

For builders and technical teams, Grok 4.3 offers serious utility at budget rates. If you are building autonomous agents that need to parse massive documents, generate functional spreadsheets, or run live code without blowing your API budget, Grok 4.3 is definitely worth testing right now.

For the AI landscape, xAI is leaning into what it has that others do not: native access to real-time X data, deeply integrated tool execution, and an ultra-aggressive pricing posture. It puts pressure on rival frontier labs that still charge premium rates for heavy reasoning workloads.

We think xAI made a pragmatic pivot here by competing on cost and agentic output formats rather than chasing benchmark bragging rights alone. The real challenge, however, will be maintaining this execution speed after losing its entire original founding research bench.