Musk pushes Tesla staff toward Grok despite admitting it lags

Elon Musk spent Friday July 10 telling Tesla engineers to switch to Grok "when possible," then spent the following day walking that back. The Electrek report, citing a memo first disclosed by The Information, has Musk pointing staff to xAI's model on the strength of lower token costs. A day later he wrote on X that he "just asked Tesla & SpaceX to try out Grok 4.5 to see if it solves their task, not use it no matter what."

Even under the softer framing, the mechanics do the pushing. Tesla capped external AI-tool spending at $200 a week for models from Anthropic, OpenAI and Google, and pointedly exempted xAI. Engineers had preferred Anthropic's Claude for development work despite months of internal beta testing of Grok under xAI product lead Andrew Milich. When your rival tools have a hard weekly ceiling and your in-house tool does not, "try it out" and "switch to it" start to look similar from the IC's seat.

The interesting part is the honesty on capability. Musk conceded that "Fable is definitely better than Grok 4.5, but most tasks don't require Fable-level capability," and the numbers back the concession: Electrek pegs Grok 4.5 at ninth overall with a 76.3 score and 68.6 on coding, the lowest in its comparison set. What tips the scale is unit economics: roughly $0.13 per task on Grok against $1.57 on Claude Fable 5. That is close to a tenth of the price, and it is the real story for any leader thinking about internal AI procurement. If most of the work your teams do is not frontier-hard, a good-enough model at a fraction of the cost is a defensible choice, even if your best engineers would rather have the best tool.

The honest caveat is that much of this rests on a leaked memo, a founder's X reply, and a benchmark ranking that will be stale in a month. What the reporting does not give you is Tesla's actual exception process, the share of engineers whose work truly needs Fable-tier capability, or how a fixed weekly ceiling holds up under agentic workloads that burn tokens fast. The move worth watching is whether other CEOs start writing similar caps, including ones without an in-house model to steer toward. Cost discipline on developer AI is coming; the Tesla version is just the one with a conflict of interest attached.