On May 2, 2026, xAI made waves in the AI industry with the launch of two groundbreaking tools: Grok 4.3, a large language model (LLM) equipped with always-on reasoning capabilities, and Custom Voices, a US-exclusive voice cloning suite. The release comes with detailed pricing, third-party benchmark results, and clear usage limitations, marking a significant step forward in xAI’s product ecosystem.
Grok 4.3: Always-On Reasoning Meets Unmatched Context
Grok 4.3 stands out from its predecessors with reasoning capabilities enabled by default—no manual toggling required. This means the model can tackle complex, multi-step tasks without additional prompting, streamlining workflows for developers and power users alike. Boasting a massive 1 million token context window, it can process extensive text and image inputs, delivering coherent text outputs. It’s fully integrated with agent workflows, allowing users to generate complex files like spreadsheets, PDFs, and PowerPoint presentations directly from prompts.
Trained on data up to December 2025, Grok 4.3 also supports real-time web searches and X platform integration, ensuring access to the latest information. Two demo cases highlight its potential: a multi-sheet Old School RuneScape (OSRS) naval combat DPS analysis tool generated in approximately 6 minutes, and a 12-page SpaceX product report complete with brand logos and data tables.
Grok 4.3 API Pricing: Reduced Costs with Tiered Structure
xAI has revamped its pricing model for Grok 4.3, offering significant cost savings compared to Grok 4.2. Input token costs are cut by 40%, while output tokens see a 60% reduction. Below is a breakdown of the API pricing structure:
| Charge Item | Fee Standard |
| Input Tokens | $1.25 per million tokens |
| Input Tokens (over 200,000) | $2.50 per million tokens (double rate) |
| Output Tokens | $2.50 per million tokens |
| Reasoning Tokens | Same as standard output token rate |
| Prompt Cache Service | $0.20 per million tokens |
| Tool Calls (Web Search, Code Execution) | $5.00 per 1,000 calls |
| Intercept Requests | $0.05 per request |
These tiered pricing options cater to both individual developers and enterprise users, scaling with usage while keeping costs manageable for high-volume tasks.
Third-Party Benchmarks: Top Performance in Niche Tasks, Room for Improvement
Independent benchmarking reveals Grok 4.3 excels in specialized domains, securing first place in the CaseLaw v2 and CorpFin benchmarks—testaments to its strength in legal and financial reasoning tasks. However, the model shows regression in two key areas: general coding tasks and agent consistency tests. Additionally, one benchmarking firm noted a 'drowsiness problem' in simulated environments, where the model may disengage or produce incomplete outputs during extended sessions.
Custom Voices: US-Exclusive Voice Cloning in 120 Seconds
Complementing Grok 4.3 is xAI’s Custom Voices suite, a voice cloning tool that creates realistic synthetic voices from just 120 seconds of audio. Currently, the service is limited to users in the United States, excluding Illinois, and is accessible via both a web-based console and enterprise API.
Pricing for Custom Voices is straightforward: text-to-speech services cost $4.20 per million characters, while the voice agent API is priced at $3.00 per hour. This makes it a cost-effective solution for businesses looking to integrate personalized voice experiences into their applications.
xAI’s latest launch combines advanced reasoning capabilities in Grok 4.3 with accessible voice cloning technology, addressing key pain points in AI workflow efficiency and personalized content creation. While Grok 4.3 has room to improve in general coding and agent consistency, its niche performance and reduced pricing make it a compelling option for targeted use cases. The US-only limitation of Custom Voices may restrict its reach, but it sets the stage for future expansion as regulatory frameworks evolve.