Musk wants Tesla employees to switch to Grok to save money, even though he admits Claude’s Fable is stronger

Musk asked Tesla employees to switch to Grok, the Grok developed in-house by his xAI, as much as possible, arguing that its token cost is only one-twelfth of Anthropic Claude Fable 5. But in a memo of his own, he admitted that Fable is clearly stronger.
(Background: Don’t invest in Musk! Two Wall Street ETFs that “isolate” Tesla and SpaceX—are the selling points political stances?)
(Additional context: The “big-name sparring” where Musk taunted Altman for “stealing from OpenAI then stealing from Apple,” and Altman hit back: SpaceX just slices retail investors)

Saving 12x in token costs comes at the price of asking employees to use a model that even the boss himself admits is stronger less—this is the contradiction Musk’s memo can’t hide. According to a report by The Information, this past Friday, Musk sent an internal memo instructing Tesla employees to “as much as possible” switch to Grok from xAI in place of third-party models such as Anthropic, OpenAI, and Google.

Set limits, but open a back door

Four days earlier—on July 6—Tesla had just set a “weekly $200” spending cap on employees’ use of third-party AI tools. The reason for the rule was straightforward: some software engineers were burning token costs of as much as several thousand dollars per week. The cap applies to models from Anthropic, OpenAI, and Google, but xAI’s Grok—and Composer, xAI’s programming tool—are explicitly excluded.

The price difference is striking. Internal data obtained by The Information shows that the token cost for Grok 4.5 to complete a task is about $0.13, while Anthropic’s Claude Fable 5 is $1.57—nearly a 12x gap. In simple terms, token cost is the compute expense consumed each time the model answers a question; a 12x difference means that the same engineering team, if it hands all work to Grok, could save a substantial budget over the course of a year.

But Musk himself also acknowledged there’s a capability gap. In his memo, he wrote: “To be fair, Fable is indeed better than Grok 4.5, but most tasks don’t require Fable-level capabilities.” That line is effectively an admission: being cheaper is the real reason behind this policy shift.

Engineers still like Claude

Sources told The Information that although Tesla has had internal tests of Grok for months, engineers generally still prefer Anthropic’s Claude in everyday development work. Besides giving instructions, Musk also required engineers to email him directly to report their experiences using Grok—something that also suggests persuading engineers to change habits isn’t easy.