Tesla employees have been urged to switch to Grok, the AI model developed by Elon Musk’s xAI division, despite internal concerns over its lower performance. According to a recent memo, Musk encouraged staff to use Grok “when possible,” highlighting its cost efficiency rather than its quality relative to competitors.
This shift follows Tesla’s recent enforcement of a weekly cap on employee spending for third-party AI tools, limiting use of products from Anthropic, OpenAI, and Google — while notably exempting Grok. The memo also asked engineers to send Musk direct feedback on Grok’s capabilities, signaling an ongoing effort to improve the tool.
Tesla has been trialing beta versions of Grok internally for several months, with the xAI product lead actively collaborating with Tesla teams to resolve functional issues. Despite these efforts, sources familiar with Tesla’s engineering workflow indicate a general preference for Anthropic’s Claude model, which remains favored for daily development tasks.
Launched alongside the acquisition of coding startup Cursor, Grok 4.5 was promoted by Musk as “Opus-class.” However, benchmark tests tell a different story: Grok 4.5 ranks ninth overall on a multi-domain leaderboard, trailing behind multiple models from OpenAI, Anthropic, and Google. Its coding accuracy rate is the lowest among tracked models, underperforming even less expensive open-source alternatives.
A noted benchmark flaw at launch involved the accidental inclusion of Cursor's own codebase in Grok’s training data, temporarily inflating one coding score, which was later corrected. On neutral coding tests, Grok scored significantly lower than Claude Fable 5, reinforcing its performance gap.
Musk acknowledged Grok’s inferiority to some competitors but argued that most Tesla tasks do not require top-tier model capabilities. The financial benefit is clear: Grok operates at roughly one-tenth the cost of the higher-performing Claude model, which likely drives Tesla’s mandate for adoption despite quality concerns.