Grok 4.5 stealth-launches in Expert mode
A post on X suggests that xAI may have stealth-launched Grok 4.5 inside the Grok application's Expert mode. Although the poster cautions that the model's self-reported version name could be the result of a web search, they note that model versions are typically hardcoded within system prompts, indicating that active testing by xAI might be underway.
Stealth-testing models in production is a common industry tactic for gathering real-world data, but LLM self-identification is notoriously unreliable due to prompt injection, search results, or hallucination.
* AI models frequently hallucinate version numbers or parrot back user leading questions.
* If valid, this stealth release suggests that an official Grok 4.5 launch is imminent.
* The lack of API documentation or official confirmation makes this speculative for now.
DISCOVERED
2h ago
2026-07-08
PUBLISHED
2h ago
2026-07-08
RELEVANCE
AUTHOR
mark_k