stark0xbt(@stark0xbt):Elon just said Grok 4.8 will have “nothing in common with 4.7 (thankfully).” It soft-launches for Grok Bot users this week. So I ran the math on the version he’s replacing. Grok 4.7 charges $6 per 1M output tokens. Claude Opus 5.5 charges $20. OpenAI’s GPT-6 Astra charges $50. That is 3.3x cheaper than Opus and 8.3x cheaper than Astra. It isn’t the smartest model. Opus scores 58 on the Intelligence Index, Grok 4.7 scores 35. But divide list price by score and Grok still wins. $0.17 per point. Opus is $0.34. Astra is $0.88. Then demand broke it. Daily users went from under 100K to over 600K in about eight weeks. About 800K pay every week. On Nov 1 Elon said serving it was “unsustainably high relative to revenue.” Limits got cut. On Nov 9, after a roundtable with the engineering team, limits were reset. Now the whole RL and inference stack has been rewritten from scratch in C/C++. Pre-training is done. 4.8 is in RL and evals are “starting to climb.” The cheapest model on the list is the one Elon is glad to leave behind. That’s the setup. All 8 numbers with the formulas are in the video. Save it.

2026.11.10 13:36

Elon just said Grok 4.8 will have “nothing in common with 4.7 (thankfully).” It soft-launches for Grok Bot users this week. So I ran the math on the version he’s replacing. Grok 4.7 charges $6 per 1M output tokens. Claude Opus 5.5 charges $20. OpenAI’s GPT-6 Astra charges $50. That is 3.3x cheaper than Opus and 8.3x cheaper than Astra. It isn’t the smartest model. Opus scores 58 on the Intelligence Index, Grok 4.7 scores 35. But divide list price by score and Grok still wins. $0.17 per point. Opus is $0.34. Astra is $0.88. Then demand broke it. Daily users went from under 100K to over 600K in about eight weeks. About 800K pay every week. On Nov 1 Elon said serving it was “unsustainably high relative to revenue.” Limits got cut. On Nov 9, after a roundtable with the engineering team, limits were reset. Now the whole RL and inference stack has been rewritten from scratch in C/C++. Pre-training is done. 4.8 is in RL and evals are “starting to climb.” The cheapest model on the list is the one Elon is glad to leave behind. That’s the setup. All 8 numbers with the formulas are in the video. Save it.

Elon Musk@elonmusk

2026.11.10 09:35

We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes. Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.

커뮤니티로 전달