ThinkFacility
  1. Home
  2. News
  3. xAI

Musk is "cautiously optimistic" SpaceX will have a Fable/GPT-6 level model in 2 to 3 months

Grok 4.7 trails Fable 5.1 by 20 points on SpaceXAI's own terminal-work test, and Musk gives SpaceX two to three months to close it.

By Think Facility · September 24, 2026 · SpaceXAI Grok

On September 24, 2026, at 11:54 am Eastern, Elon Musk put a date on SpaceX's next big model. Replying to a user on X, he wrote that he's "cautiously optimistic that SpaceX will have a Fable/GPT-6 level model in 2 to 3 months."

Fable 5.1 and GPT-6 Astra are the two models Anthropic lined its own Opus 5.5 up against at launch. SpaceXAI released Grok 4.7 three days before Musk's post, on September 21. What surprised me is the order he gave: two to three months to catch up, then about six to lead. He didn't name the model.

Claim
a Fable/GPT-6 level model in 2 to 3 months
Longer goal
SpaceX will reach pole position in about 6 months
Latest release
Grok 4.7, Sep 21, 2026
Terminal-Bench 4.0
Grok 4.7 37.6%, Fable 5.1 Max 57.9%

How the thread went

The post he answered came from @notjazii, who called Grok 4.7 "the biggest fumble" and still had high hopes for Grok 4.8. Half an hour after Musk replied, another user, @vasalex93, wrote that Anthropic and OpenAI would have something better by then anyway.

Elon Musk in a plaid shirt on stage before a lit tunnel mouth, a slide on a screen beside him and phones raised in the crowd
Elon Musk presenting at a Boring Company event, in front of one of its tunnels. Photo: Steve Jurvetson from Menlo Park, USA, CC BY 2.0, via Wikimedia Commons
  1. I fan post
  2. Musk Musk: 2 to 3 months
  3. Alex the doubter
  4. Musk Musk answers

Sep 24, Eastern time, read off each post on X.

Musk's answer to that doubter is the longer post, and it's where the six months comes from.

Elon Musk@elonmusk · xAI

1. We will keep accelerating. Our AI efforts are only 3 years old, vs 6 and 10 years old for Anthropic and OpenAI. If our second derivative remains strong, SpaceX will reach pole position in about 6 months. 2. Once you far exceed the caliber of intelligence needed for a class of tasks, additional intelligence is pointless. You don’t need (and it would be cruel to put) Newton-level intelligence in your toaster! 3. Hardware is hard. Bringing massive compute online rapidly is incredibly difficult. SpaceX has demonstrated exceptional ability in this regard and will only get better.

View the post on X

His second point is the one aimed at @vasalex93. Once a model is far past what a class of tasks needs, Musk says, more intelligence is pointless (hence the toaster).

How far back Grok 4.7 sits

SpaceXAI's own launch table shows the size of the gap. On Terminal-Bench 4.0, a test of multi-hour terminal work, Grok 4.7 scored 37.6% to Fable 5.1 Max's 57.9%. On CursorBench 4.0 it's 46.3% against 51.8%. Grok 4.7 did come out ahead on EEBench and the Harvey legal benchmark, and it costs $2 per million input tokens where Fable costs $10. We went through that table when it came out.

It works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date.

From Introducing Grok 4.7 | SpaceXAI

The last date he gave

Musk has put dates on SpaceX models before this month. On September 13 he posted that Grok 4.8, "a 2.5T model trained with our new C++ software stack," would finish training that week and start RL. We wrote that one up at the time. Neither of his September 24 posts says where Grok 4.8 stands now, or whether it's the model he means.

His third point was about hardware, and he came back to it later that day. At 2:22 pm Eastern he posted that "the amount of compute in space will obviously round up to 100% of all compute."

More on xAI

All xAI stories