SpaceXAI: Grok 4.3 (batch)

Grok 4.3 (batch) is a reasoning model from xAI, listed on api14 under the model id x-ai/grok-4.3:batch as SpaceXAI: Grok 4.3 (batch). It sits in the text category and is offered as a batch variant, which means it is intended for workloads that can be submitted and processed together rather than one interactive request at a time. The provider describes it as a reasoning model suited to agentic workflows, instruction-following tasks, and applications that need high factual accuracy.

The model accepts text, image, and file inputs and returns text. Its context window is 1,000,000 tokens, and the maximum output is 900,000 tokens, so very large inputs and very long generations are both possible within a single call. In addition to standard text generation, it exposes reasoning-oriented parameters, so you can control how much reasoning effort is applied and read reasoning content alongside the final answer.

Those capabilities point to a fairly specific set of uses. Because it takes images and files as well as text, it fits document and mixed-media analysis where the source material is long: contracts, reports, transcripts, logs, or scanned material that needs to be read and reasoned over in one pass. The large context and output limits make it a reasonable fit for agentic pipelines that accumulate state over many steps, for instruction-following tasks with detailed specifications, and for structured extraction where the answer must come back in a defined shape. Batch processing suits jobs that are not latency sensitive, such as bulk classification, large scale summarisation, or evaluation runs over many records.

You call it through api14's OpenAI-compatible API using the model id x-ai/grok-4.3:batch, so existing OpenAI client libraries and request shapes work with a changed base URL and model name. The supported request parameters are include_reasoning, logprobs, max_tokens, reasoning, reasoning_effort, response_format, seed, structured_outputs, temperature, tool_choice, tools, top_logprobs, and top_p. Tool calling and structured outputs are available, which lets you wire the model into function-calling agents or force responses into a schema via response_format. Logprobs and top_logprobs are supported if you need token level probabilities, and seed plus temperature and top_p give you control over reproducibility and sampling.

Since this is the batch variant, it is the better choice when throughput matters more than per-request latency; if you need immediate interactive responses, a non-batch version would generally fit better. Its inputs are text, image, and file, and its output is text only, so it is not the model for generating images, audio, or other media. If your task is small and simple, the million token context and reasoning controls may be more than you need, but for long-context, tool-using, structured work, the supported parameters cover most of what a developer would reach for.