xAI — Grok model development¶

Development line: organization:xai · thread grok-model-development
Events: 1 dated event, 2025-07-08 → 2025-07-08 · Researched: 2026-09-04 · confidence: medium

What it is¶

xAI makes Grok and provides an API for teams building chat, coding, and agent workflows.

Capabilities:

• Grok 4.6 accepts text and images and returns text;
• Grok 4.6 supports configurable reasoning;
• Grok supports function calling, Web/X Search, and code execution;
• Voice and Imagine APIs are separate.

Limit: Grok 4.6 has a 500k-token context window; below 200k input tokens, its API price is $2 / $0.50 / $6 per 1M input / cached-input / output tokens.

For a new text or coding integration, use the current model rather than the original Grok 4 slug.

Development line¶

• 2025-07-08 — xAI’s Grok 4 development milestone. Grok 4 and Grok 4 Heavy reached subscribers and the xAI API with native tool use and real-time Web/X search.

What changed¶

2024-08-15 — “uncensored” is an unverified positioning label. No first-party release record connects it to a Grok version, endpoint, policy, or availability change.

2025-07-08 — The linked social post was not retrievable. The first confirmed product change is 2025-07-09: Grok 4 and Grok 4 Heavy reached subscribers and the xAI API with native tool use and real-time Web/X search.

2026-05-15 — grok-4-0709 and several older Grok slugs were retired; calls to the old Grok 4 slug redirect to grok-4.3.

As of 2026-09-04, Grok 4.6 has been available in the API since 2026-08-12 and is the current default for code and general text work.

How to use this¶

As of 2025-07-08, practitioners evaluating xAI’s Grok should revisit model-selection assumptions at the Grok 4 milestone. Verify current capability and access details from primary sources before adopting it.

1. Create an xAI Console account and add credits. — https://docs.x.ai/developers/quickstart
2. Create an API key and load it through XAI_API_KEY. — https://docs.x.ai/developers/quickstart
3. Select grok-4.6 for a new general text or coding integration; use a dated slug when output consistency matters. — https://docs.x.ai/developers/models
4. Send the first request to the Responses API with the xAI SDK or an OpenAI client pointed at https://api.x.ai/v1. — https://docs.x.ai/developers/quickstart
5. Enable Web Search or X Search when an answer depends on current information. — https://docs.x.ai/developers/models

Best practices¶

• Pin a dated model slug for reproducible production behavior. Use -latest only when automatic model updates are wanted. — https://docs.x.ai/developers/models
• Treat current-event questions as a tool-use case. Enable server-side search rather than relying on the model’s training cutoff. — https://docs.x.ai/developers/models
• For repeated conversations, use a stable prompt_cache_key or x-grok-conv-id. Keep the prefix append-only and monitor cached_tokens. — https://docs.x.ai/developers/advanced-api-usage/prompt-caching/best-practices
• Budget against both RPS and TPM. Handle HTTP 429 with bounded exponential backoff. — https://docs.x.ai/developers/rate-limits

Superseded by this¶

• 2026-05-15 — The 2025-era grok-4-0709 deployment target is obsolete: xAI retires it and redirects calls to grok-4.3; choose a replacement explicitly to control reasoning and price.
• 2026-08-12 — Default guidance for new general text and coding work moved to grok-4.6; this does not itself retire supported grok-4.3.

Still unknown¶

• The 2024-08-15 reference is an X search URL, not an archived release note; it did not establish the claimed “uncensored” state.
• The 2025-07-08 x.com status URL was inaccessible at review time, so its wording was not used.
• No first-party Simplified Chinese model-specific release or documentation page was found. The fact-bearing sources are English.

Sources¶

Agent brief¶

• Subject: organization:xai, thread grok-model-development, 1 dated event 2025-07-08 → 2025-07-08.
• Practical note: As of 2025-07-08, practitioners evaluating xAI’s Grok should revisit model-selection assumptions at the Grok 4 milestone. Verify current capability and access details from primary sources before adopting it.
• Confidence: medium. The dated superseded entries above are the authority for what is obsolete.