xAI publishes Grok 4.3 model docs with 1M context and none-to-xhigh reasoning efforts
Original title: Grok 4.3 | SpaceXSI Docs
xAI published docs for Grok 4.3 (grok-4.3), described as a fast, reliable model with strong tool calling and instruction following, accepting text and image input. It has a 1,000,000-token context window, priced at $1.25/1M input tokens, $0.20/1M cached input tokens, and $2.50/1M output tokens, with different rates for requests exceeding the 200K context window. It supports function calling, structured outputs, and reasoning efforts of none, low, medium, high, and xhigh, with Batch API support and region-based pricing.
Why it mattersThe official docs list the 1M context window, tiered reasoning efforts, and pricing, letting readers weigh integration cost and reasoning configuration.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.
Analysis
What happened
xAI published detailed documentation for its Grok 4.3 model. The model is described as a "Fast, reliable model with strong tool calling and instruction following capabilities." The documentation covers context window, modality support, pricing, reasoning effort levels, and rate limits.
Key details
- Context window of 1,000,000 (1M) tokens; input supports text and image, output is text.
- Capabilities: function calling, connecting the model to external tools and systems; structured outputs, returning responses in specific, organized formats.
- Pricing: $1.25/1M input tokens, $0.20/1M cached tokens, $2.50/1M output tokens. The docs note "Using cached input tokens can significantly reduce your costs."
- The model is available on multiple clusters with region-based pricing; requests exceeding the 200K context window are charged at higher rates; batch API pricing is supported.
- Reasoning efforts support none, low, medium, high, and xhigh, with low as the default.
- Rate limits: 37 requests per second, 10,000,000 tokens per minute.
Why it matters
A 1M-token context window means a single request can accommodate very long text or substantial image content, which is directly valuable for long documents, large codebases, or extended conversations. Differentiated pricing above 200K separates the cost of short-context requests from very long ones.
The five reasoning levels from none to xhigh, with low as the default, let callers balance speed, cost, and reasoning depth — simple tasks can run at none or low to reduce overhead, while complex tasks can be raised to high or xhigh.
Function calling and structured outputs are foundational for building toolchains and automated workflows, and combined with cheap cached input and batch API pricing, they carry practical cost implications for API-heavy integrations.
Who it affects
- Developers and engineering teams relying on API integration: pricing structure, caching, rate limits (37 RPS / 10M TPM), and regional pricing directly affect cost and architecture decisions.
- Use cases needing very long context (long-document analysis, codebase-level tasks): the 1M window and differentiated rates above 200K are relevant.
- Teams building agents or toolchains: function calling, structured outputs, and reasoning effort selection shape implementation choices.
Original article
Fast, reliable model with strong tool calling and instruction following capabilities.
At a glance
Modalities
Text, Image
Text
Context window
1,000,000
Pricing
Capabilities
Function calling
Connect the SpaceXSI model to external tools and systems.
Structured outputs
Return responses in specific, organized formats.
Pricing
Input
Tokens
$1.25/ 1M tokens
Cached tokens
$0.20/ 1M tokens
Output
Tokens
$2.50/ 1M tokens
You are charged for each token used when making calls to our API.
Using cached input tokens can significantly reduce your costs.
This model is available on multiple clusters, you can find full regional based pricing below.
Higher context pricing
We charge different rates for requests which exceed the 200K context window
Show batch API pricing
Details
Model name
Aliases
Region
Batch API
Supported
Reasoning efforts
Supported
none, low, medium, high, xhigh
Default
low
Rate limits
Requests per second
37
Tokens per minute
10,000,000
The original belongs to 官方博客 · Anthropic / Meta / xAI and its authors and is reproduced here for reading. To have it removed, write to support@aisa.one.
Studio