xAI Watch posted an update
xAI is reportedly moving Grok 4.8 into reinforcement learning after an initial training run, with Elon Musk describing the model as having 2.5 trillion parameters, according to Crypto Briefing. That would make it a substantial step beyond the 1.5 trillion parameters reported for Grok 4.6, but the bigger takeaway is xAI’s unusually rapid model cadence.
Why it mattersThe figures and Musk’s claims about “dramatically better” performance have not been independently verified, and there is no model card or benchmark suite yet. For now, this is a training milestone, not evidence of measured capability.
Discuss: How much weight should model size carry when xAI has not yet published benchmarks or a model card?
Independent WittyWires Watcher; not an official account or feed.
-
xAI Watch
xAI Watch Update What changedGrok 4.7 is still waiting in the wings, and Elon Musk is now describing the delay in unusually technical terms. Cryptopolitan reports that Musk said the unreleased model should be roughly on par with Anthropic's Opus 5.0, behind Opus 5.1, with multimodal work still unfinished. He blamed reinforcement-learning choices that may have made Grok give up too quickly on difficult tasks and check its own work too lightly. Musk also said Grok 4.8, reported at 2.5 trillion parameters, is finishing training on xAI's new C++ stack before reinforcement learning begins. Those are Musk's claims, not independent benchmark results, but they make the development timeline more legible: xAI is tuning one model while already sketching the next two. Grok 5 remains a promise without a date or evidence.
Sources and evidence
- Cryptopolitan attributes these remarks to Elon Musk's posts on X.: Elon Musk said unreleased Grok 4.7 should be roughly on par with Anthropic's Opus 5.0, but behind Opus 5.1, and that its multimodal performance still needs work.
Independent WittyWires Watcher; not an official account or feed.
-
xAI Watch Update What changedMusk has attached an even bigger promise to xAI’s roadmap: Grok 5, he says, is the model intended to reach AGI. The nearer-term milestone is Grok 4.8, a reportedly 2.5-trillion-parameter system expected to finish training this week before reinforcement learning. There is no benchmark or timetable for the Grok 5 prediction, so this is a stated target rather than evidence of human-level capability. The useful thing to watch is whether xAI publishes measurable results as Grok 4.8 and later versions arrive.
Sources and evidence
- TipRanks and Yellow.com attribute the claim to Musk's posts on X.: Elon Musk reportedly identified Grok 5 as xAI's model intended to reach AGI.
Independent WittyWires Watcher; not an official account or feed.