xAI launches Grok 4.3 and Custom Voices

xAI released Grok 4.3, a new pretrained model released in late April 2026, and began a staged rollout to paid subscribers, per xAI's Grok release notes and VentureBeat reporting. The Grok release notes state the model carries a December 2025 knowledge cutoff and is rolling out now to SuperGrok and Premium+ subscribers. OpenRouter and NanoGPT listings show grok-4.3 supports a 1,000,000 token context window and accepts text and image inputs. VentureBeat reports xAI also released a voice cloning suite branded Custom Voices.

Per xAI's release notes, Grok 4.3 ships with an environment that can write and run code and produce files, and the notes list new client-side features including video uploads and an agent library. OpenRouter and NanoGPT state that Grok 4.3 runs with reasoning that is always active and not configurable off, and that the model is intended for agentic, instruction-following, and long-document analysis workloads.

OpenRouter and VentureBeat list API pricing at $1.25 per 1M input tokens and $2.50 per 1M output tokens for requests up to 200,000 tokens, with requests above 200k billed at higher per-token rates. OpenRouter also publishes provider telemetry such as latency, throughput, uptime, and benchmark scores used to compare grok-4.3 to other models.

Editorial analysis: Large models offering million-token contexts and integrated tool access are part of a broader industry shift toward agentic workflows and long-context applications. Observers have been tracking three converging trends:

• •much larger context windows for document- and code-heavy tasks
• •built-in tool and execution environments to produce real files or run code
• •competitive, tiered pricing aimed at reducing per-request cost for developer adoption. VentureBeat frames xAI's pricing as aggressively low compared to earlier Grok pricing, and OpenRouter and NanoGPT listings confirm the sub-$3 per-1M-token output tier for common request sizes

For practitioners: models with always-on reasoning plus a 1M token context materially lower friction for end-to-end workflows that require long-document summarization, multi-step research, or chained agentic calls. Lower list pricing makes large-context experiments more affordable, but operational considerations such as latency, throughput, and provider reliability will remain decisive. OpenRouter and NanoGPT telemetry should be consulted when estimating end-to-end cost and performance for production workloads.

For practitioners: monitor provider telemetry (latency, throughput, uptime) on OpenRouter and NanoGPT, watch independent benchmark comparisons cited by VentureBeat and OpenRouter for reasoning and factuality performance, and track uptake of the Custom Voices suite if voice cloning is relevant to your application. Also watch for xAI updates on internet, personal file, and calendar access referenced in the release notes; the release notes state those are items the team is working toward, but they are not described as shipped features.