- Published on
Musk tells Tesla staff to switch to Grok — a model he admits is worse
- Authors
- Name
- DeepHorus
Elon Musk has directed Tesla employees to adopt Grok, the AI model from his own xAI, despite acknowledging it underperforms competitors. In a Friday memo first reported by The Information, Musk urged staff to switch to Grok 4.5 when possible, citing lower token costs. The directive follows Tesla capping third-party AI spending at $200 weekly, a limit that exempts Grok.
Tesla has been testing Grok beta versions internally for months, with xAI product lead Andrew Milich working alongside Tesla staff. However, four people familiar with internal usage revealed that engineers broadly prefer Anthropic’s Claude for day-to-day development. The memo now asks engineers to email feedback directly to Musk, signaling a top-down push for adoption.
Grok 4.5, released Wednesday jointly with Cursor, ranks 9th on an aggregate multi-domain leaderboard with a score of 76.3. It trails behind multiple models from OpenAI, Anthropic, and Google, including Claude Fable 5 and GPT-5.6 Sol. Its coding score of 68.6 is the lowest on the board, and on the DeepSWE 1.1 benchmark, Grok scored 53% versus Claude Fable 5's 70%.
The launch was also marred by a benchmark controversy: Cursor disclosed that an earlier snapshot of its own codebase was accidentally included in Grok 4.5’s training data, inflating one headline coding score. Musk did not dispute the gap, writing on X, “In fairness, Fable is definitely better than Grok 4.5, but most tasks don’t require Fable-level capability.” He argued Grok’s advantage is price—$0.13 per task versus $1.57 for Claude Fable 5.
The directive raises governance concerns for Tesla, a publicly traded company. Critics call it self-dealing, as Musk owns xAI (now folded into SpaceX) and benefits from steering Tesla to his own product. The $200 cap that conveniently exempts Grok further highlights potential conflicts of interest, especially since engineers previously chose Claude on merit.
Some argue that the real solution is self-hosting open-weight models like DeepSeek-V4 and GLM-5.2, which deliver 90-95% of frontier capability at a fraction of the cost. This approach could achieve cost savings without mandating an inferior product. With the benchmark data and internal preferences, Tesla engineers may face productivity trade-offs if forced onto Grok.
The move underscores the risks when a CEO’s multiple roles intersect with corporate decision-making. Whether Tesla will persist with enforcing Grok or allow engineers latitude remains uncertain, as the company balances cost control against employee satisfaction and performance.
