Grok Bot fixes, Grok 4.8 and Musk on Opus 5.5

Grok Bot reset weekly limits on November 9 after promised fixes, Grok 4.8 heads to Bot users this week, and Musk called Opus 5.5 better than Grok 4.8.

This week in Grok: the Grok Bot team promised fixes for usage limits and model routing on November 8 and reset everyone's weekly limits on November 9 · Grok 4.8 is in its last training stage, with a soft rollout to Grok Bot users planned for this week · Elon Musk said Grok 4.8 is "not as good" as Anthropic's Opus 5.5 and that Grok 5 should pass it in two or three months. Below is what was said, who said it, when, and what it means if you run Grok Bots.

Most of this week's news comes from posts on X by Musk and the Grok Bot team rather than from official xAI announcements. We link every post we quote so you can read it in full, and we say clearly where a detail comes from news reports instead.

The short version

Grok Bot had a rough fortnight. Around Halloween, many subscribers saw their weekly allowance jump to 100 percent, and on November 1 Musk said the cost of serving Grok Bot had been too high for the revenue it brought in. A week later, on the evening of November 8 (US time), Musk said he had reviewed the product with the Grok Bot engineering team and that fixes were going out overnight, and the next day the official Grok Bot account reset weekly limits for every user. On November 10, he said Grok 4.8, the next model, was in reinforcement learning and would reach Bot users first. The same day, answering a question on X, he gave a blunt assessment: Anthropic's Opus 5.5 is better than Grok 4.8 today.

How the usage limit problem started

Over the weekend of October 31 and November 1, 2026, Grok Bot users began reporting that their weekly usage meters had filled far faster than normal. TechCrunch reported that some users saw usage jump from as low as 50 percent to 100 percent, including users who said they had not run their agents overnight.

Musk confirmed it was not a glitch. Replying on X on November 1, he wrote: "The cost of serving Grok Bot was unsustainably high relative to revenue." He did not publish new limit numbers or announce a reset, and neither did xAI. We covered that exchange in Grok Bot usage limits: Musk says serving costs were too high.

Teslanorth later quoted one subscriber who said a $200 plan that used to last a week of heavy use ran out after about 14 hours of moderate work. That is a single user's account and was not independently checked, but it matches the tone of the replies under Musk's posts that week.

November 8: "Fixes are being implemented overnight"

On the evening of Sunday, November 8 (02:25 UTC on November 9), Musk posted: "Just did a round table review with the Grok @Bot engineering team. Fixes are being implemented overnight and you should notice a dramatic improvement by tomorrow. If not, reply to this post and we will address any remaining issues right away."

The post did not say what was being fixed. That detail came 25 minutes later from Marco d'Amore, whose X profile says he is building Cursor, one of the companies that sells Grok Bot on its plans. He quoted Musk's post and wrote: "We hear all your feedback, especially on usage limits and model routing and will be improving things overnight." He added that "over the coming weeks many improvements to the harness and routines will further enable you to get much more out of each subscription," and that the Bot team wants to deliver the product "in a way that is also sustainable."

Read together, the two posts name three areas: usage limits, model routing, and the harness and routines that run your Bots. Neither post promised higher allowances, a refund or a price change.

November 9: weekly limits reset for everyone

On November 9 at 19:17 UTC, the official Grok Bot account on X posted: "We've reset weekly limits for all Grok Bot users. Enjoy!" That gave every subscriber a fresh weekly allowance, including people who had been locked out since the Halloween weekend.

It was a one time reset, not a bigger allowance. The post said nothing about raising weekly limits or changing prices, and as of November 11 xAI has not published new limit numbers. If your routines run the same way they did in late October, you can expect to hit the same ceiling again unless the routing and harness changes make each run cheaper.

What "model routing" means for your Bots

The posts did not define the term, so this is our explanation, not xAI's. In a product like Grok Bot, routing usually means deciding which model handles each step of a job. A quick check of an inbox or a status page does not need the largest model, while planning a multistep research task might. If simple steps are sent to an expensive model, a routine burns through an allowance much faster than its owner expects.

That would explain why usage limits and routing were mentioned in the same sentence. Better routing can lower the cost of each run without changing the size of anyone's allowance, which fits Musk's earlier point that the problem was serving cost rather than demand. The mention of the "harness" points the same way: it is the layer that manages a Bot's context, tools and steps between model calls, and a leaner harness reads less context again on every run.

Grok 4.8: rebuilt from scratch, rolling out to Bot users first

On November 10, Musk gave the most detailed update yet on Grok 4.8. He wrote that xAI had "dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time." Pretraining is finished, he said, a lot of cleanup was needed on the supervised fine tuning data, and "now, 4.8 is in RL and the evals are starting to climb."

The part that matters most for Grok Bot users: "We will do a soft rollout to Bot users later this week and see how it goes." In the same post he wrote that Grok 4.8 "will bear no resemblance to 4.7 (thankfully)" and that it "is not yet incredible, but, this time, the foundations are built right." In a follow up about the new code he added: "One day, super intelligence will write a better software stack, but that day has not yet come. This is tight code."

A soft rollout normally means a share of users get the new model first while the team watches quality, speed and cost before widening access. Starting with Grok Bot makes sense given this month's cost problems. Bots run long, repeated jobs, so they show quickly whether a new model and a new inference stack are cheaper to serve. The full history, including the dates Musk gave earlier this year, is in Grok 4.8 and Grok 5 release dates.

Musk on Opus 5.5: "a masterpiece"

Under the same thread, a user asked how Grok 4.8 compares with Anthropic's Opus 5.5. Musk replied: "Not as good. Opus 5.5 is a masterpiece. But Grok 5 should exceed Opus 5.5 and maybe that happens in 2 or 3 months, with continued acceleration thereafter."

It is unusual for the head of an AI lab to say a rival's model is better than the one he is about to ship. Outlets including IBTimes UK, Yahoo Tech and FourWeekMBA picked up the comment. Two cautions apply. First, Musk gave no benchmark scores, so "not as good" is his judgment, not a published measurement. Second, the Grok 5 window is hedged with "should" and "maybe", and earlier Grok 5 targets have passed without a release. Two or three months from November 10 would put it somewhere between mid January and mid February 2027, if it holds.

Also this month: changes for API users

Developers had two smaller changes, both listed in the official Grok API release notes. In November, the grok-imagine-image-quality model was retired: requests to it, and to grok-imagine-image-pro, are now served by grok-imagine-image-2.0 at its low quality setting and a lower price per image, and xAI says no request changes are needed. In October, text API requests gained a guidelines field for custom moderation, which lets an app set how Grok handles adult content, violence and profanity.

Grok Voice 3.0, released in October, remains the current speech to speech model at the same price as Grok Voice 2.0. We covered it in Grok Voice 3.0.

What to do if you run Grok Bots this week

Check your usage after the reset. In the Grok Bot desktop app, open Settings, then Usage & Billing. The Grok Bot usage section charts your usage by Bot or by source, and Usage limits shows how much of your weekly allowance is gone and when it resets. If your meter still fills faster than your Bots' workload explains, Musk asked people to reply to his November 8 post. A reply with your plan, the routine that ran and how far the meter moved is more useful to the team than a general complaint.

Run routines only as often as the job needs. A daily briefing rarely needs to run every hour, and a weekly report does not need a daily check. Fewer runs is still the simplest way to stretch an allowance, whatever changes in routing.

Trim what each Bot carries. Long saved context is read again on every run. The shared Bot trimmy is built for exactly this, and our guide on using Grok Bots safely covers keeping each Bot's access narrow.

Expect a different model mid week. If you are in the Grok 4.8 soft rollout, a routine that worked well on 4.7 may behave differently. Musk himself says 4.8 bears no resemblance to 4.7. Check the first few runs of anything that sends messages, edits files or touches money before letting it run unattended again.

Give each Bot one job. Bots with a single clear purpose are easier to route cheaply and easier to check. If you are looking for focused Bots, browse Top Grok Bots by job.

What we are watching next

Three things will tell us whether this week's promises held. First, whether xAI publishes clear weekly limit numbers for each plan. The usage screen shows your own share used, but not how large the allowance is in tokens or runs. Second, whether Grok 4.8 actually reaches Bot users this week, and whether xAI follows it with an official announcement, a model card or API availability. Third, what the "many improvements to the harness and routines" turn out to be over the coming weeks. We will add each one to the Grok timeline as it happens.

Questions people are asking

Did xAI raise Grok Bot usage limits? No. It reset everyone's weekly limits once, on November 9, and promised fixes to usage limits and model routing. No higher allowance, refund or price change was announced.

Is Grok 4.8 out? Not as of November 11. Musk said a soft rollout to Bot users would start later in the week of November 10. There is no xAI announcement, model card or API listing for it yet.

Did Musk really say Opus 5.5 is better than Grok? Yes, about Grok 4.8. On November 10 he replied "Not as good. Opus 5.5 is a masterpiece," and said Grok 5 should exceed it, maybe in two or three months.

When is Grok 5 coming out? There is no date. Musk's latest hint is two or three months from November 10, hedged with "should" and "maybe".

Who is Marco d'Amore? His X profile says he is building Cursor, which offers Grok Bot on its plans. His post quoting Musk's is the only one that names usage limits and model routing as the areas being fixed.