Farzad: Grok Bot "SIGNIFICANTLY improved for me" + Fable 5.2 "20% cheaper"

Farzad says Grok Bot "SIGNIFICANTLY improved for me as well this AM," and by Monday afternoon he was posting "We're so back." That's his experience. Joey Gilkey, who met the Grok Bot team, says the team admitted the weekend's problems came almost entirely from pulling Opus 5.5 out. Plus Fable 5.2 and Grok at SpaceX.

"SIGNIFICANTLY improved for me as well"

Yesterday's Digest covered Grok Bot's rough weekend. Monday's reports went the other way.

Mike P (mikepat711), Monday at 11:03am CST:

My Grok Bot morning routines were pristine today. No amnesia or strangeness. Will keep an eye out for problems but so far nothing I was seeing last week.

Man I love this product.

Farzad quoted it at 11:23am CST:

Grok Bot SIGNIFICANTLY improved for me as well this AM.

That's his experience, and it's the whole post. He doesn't say which workflows he ran or what changed. At pull time it had 430 likes, 24 replies and 12,156 impressions.

At 1:17pm CST the @bot account posted "We've reset weekly limits for all Grok Bot users." and "Enjoy!" Farzad liked it. At pull time it had 35,898 likes, 4,126 replies and about 1.64 million impressions. Eleven minutes later, at 1:28pm CST, he quoted it. His post reads "Fable 5.2." then "Grok Bot fixed with weekly reset." then a Jensen Huang line this Digest skips, then "We're so back." It drew 401 likes. "Fixed" is his call. The @bot post announces a limit reset and doesn't say what was wrong.

Emily Gavrilenko (egavrilenko11), whose X bio reads "building @spacexai," quoted the reset at 1:36pm CST:

we've heard your feedback and reset @bot limits

the team is sprinting this week to make grok bot usage go farther and improve the product experience

let us know what we should focus on first

Farzad liked it and RTd it at 3:35pm CST. Emily doesn't say what the sprint will change.

Check Mike P here. Farzad here. @bot here. Farzad's "We're so back" here. Emily here. Farzad's RT here.

"Almost entirely from pulling Opus 5.5 out"

Joey Gilkey (JoeyGilkey) posted "Update #2 (after meeting with the @bot @spacexai engineering team)" Monday at 3:36pm CST. The first three points:

• Team is great. They care a ton. They owned it.
• They admitted that it was almost entirely from pulling Opus 5.5 out. Which is now back as of this AM.
• Learned a little bit about the context-compaction process with Grok 4.7-summarize which is not a public model. Plays a role in routing config and the summary we get back.

That's Joey's account of what the team told Joey. Nothing from the @bot account in this pull confirms it.

Joey is upfront that the rest is partly a guess, writing "I am somewhat surmising on why stuff went so poorly over the weekend." The guess is that work in progress was likely leaning heavily on Opus 5.5, "then it got pulled, routed to different model, forgot the context, and turned incredibly dumb." Joey writes, "We laughed when I showed them a project that was mid flight in a doc," with a first half that was "intelligently designed, formatted" and a second half that was "a wall of seemingly incoherent, non formatted text." Joey repeats the caveat: "Again, I'm surmising but I think it was dumber model with all context removed". The update also says, "The problems we all experienced should be fixed."

The update quotes Joey's own Sunday 2:43pm CST post: "We were slated to sign for a mid 6-figure contract this week with @spacexai to roll out @bot across the org." That post says the deal "fell apart over the last few days with the severe drop in quality with no control over models and the comments on it being unsustainable," and adds, "Had a killer pilot with them but it died at the finish line." On Monday, Joey is "Honestly still weighing the decision vs just using team plan."

Farzad liked the update and RTd it at 4:46pm CST. At pull time it had 1,784 likes, 482 bookmarks and about 168,000 impressions.

Check Joey here. Joey's Sunday post here. Farzad's RT here.

"Usage by source, as well as usage by Bot"

Sawyer Merritt, Tuesday at 7:37am CST:

A new feature has been added in Grok @Bot's "Manage usage" settings. It now gives you a breakdown of usage by source, as well as usage by Bot. It updates on a daily basis.

That's Sawyer's description, with an image attached. Nothing from the @bot account in this pull announces the feature, and Sawyer's text doesn't say what counts as a "source." Farzad liked it and replied at 7:43am CST: "Super duper helpful actually." Sawyer answered at 7:46am CST, "Was very much needed," and Farzad liked that too. At pull time the post had 477 likes and about 30,000 impressions.

Check Sawyer here. Farzad's reply here. Sawyer's answer here.

"20% cheaper vs previous Fable"

The Claude account (claudeai) introduced Claude Fable 5.2 and Claude Mythos 5.2 Monday at 12:57pm CST, saying they "show improvements in areas like coding, reasoning, and vision." Farzad liked it. Six seconds later the account posted pricing:

Claude Fable 5.2 is available today on all paid plans, on the Claude Platform, and on all major cloud platforms.

It's $8 per million input tokens and $40 per million output tokens, down from $10 and $50.

Farzad quoted that at 1:10pm CST: "Fable 5.2 20% cheaper vs previous Fable." The math works for both prices: $8 is 20% below $10, and $40 is 20% below $50. Claude's post says "down from" without naming the earlier model, and the 20% is Farzad's math. Anthropic's pricing page lists the previous Fable, Claude Fable 5.1, at $10 and $50, and Fable 5.2 at $8 and $40. His post had 132 likes.

Matthew Berman (MatthewBerman), 1:14pm CST, which Farzad liked:

I've burned over 100 billion tokens early testing Fable 5.2. It is BY FAR the best model I've ever used.

Berman's review calls it "the most autonomous model to date" and says that, given little direction, "it makes assumptions that 99 times out of 100 are exactly what I would have done." The review says it's "so good at controlling Unreal Engine and Blender" and attaches what Berman calls a Gears of War clone. Those claims come from Berman's own testing. The review ends with a catch: "I will likely still prefer Opus because I can run Opus nearly 24/7 and rarely run out of quota. And for the hardest jobs, I'll switch to Fable 5.2 for planning." Berman names that model as Opus 5.5, the model Joey Gilkey named above. Berman's post had 2,057 likes.

Check Claude here. Claude on pricing here. Farzad here. Matthew Berman here.

"Grok analyzing the same video is 99.9% perfect"

Sawyer Merritt, Monday at 12:18pm CST, which Farzad liked:

SpaceX VP of Launch @TurkeyBeaver says Grok is now helping them identify foreign object debris in aerospace parts, and that Grok (99.9%) is better at it than humans (94%).

"We've automated being able to detect FOD. We'll have an inspector go in and borescope a tube, and we find that humans are about 94% perfect (at) finding a piece of it. Grok analyzing the same video is 99.9% perfect."

The numbers come from the VP's quote as Sawyer relays it. The full article Sawyer links is an Ars Technica piece by Eric Berger, published November 9, which names the speaker as Kiko Dontchev, SpaceX's vice president of launch, speaking at the UP.Summit conference in Florida the week before. It quotes the same 94 percent and 99.9 percent figures and reports Dontchev saying FOD detection has been largely automated using Grok. The quote describes one task, an inspector borescoping a tube, with humans and Grok reviewing the same video. It doesn't give a sample size or say how "perfect" was scored. At pull time the post had 1,967 likes and 883,644 impressions.

Check Sawyer here.

"We have dumped JAX/SGLang"

Monday at 6:23pm CST, cradrrat asked: "Can we have an update on when to expect Grok 4.8 please @elonmusk". Elon answered Tuesday at 3:35am CST:

We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.

He says "Training (aka pre-training) is done," that "4.8 is in RL and the evals are starting to climb," and that "We will do a soft rollout to Bot users later this week and see how it goes." He says Grok 4.8 "will bear no resemblance to 4.7 (thankfully)" and adds, "It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special." The rollout timing and "something special" are Elon's forecasts. He gives no eval numbers.

Farzad liked it and quote-posted it at 7:21am CST, asking @grok to explain the thread. At pull time Elon's reply had about 17,500 likes and about 1.66 million impressions.

At 4:34am CST, about an hour after the 4.8 update, Elon posted:

Turns out that the Starlink software, built for managing the movement of vast amounts of data through space and to the ground, is also very well suited to training super intelligence

Farzad liked it and RTd it at 6:58am CST. That's Elon's claim. He doesn't name the software or say whether it ties to the rewrite. It had about 44,600 likes, 2,785 replies and about 2.5 million impressions.

Check cradrrat here. Elon on Grok 4.8 here. Farzad's quote here. Elon on Starlink here. Farzad's RT here.

"Now live in Ohio"

Lemonade (Lemonade_Inc), Monday at 11:08am CST:

Lemonade Autonomous Car is now live in Ohio.

Tesla drivers can save 50% on miles driven with FSD (Supervised), and 30% with FSD v14 Lite.

Farzad RTd it at 11:23am CST without comment. Those are Lemonade's numbers, and as worded both discounts apply to miles driven on FSD. The link preview's headline reads "50% Off Tesla FSD Miles." The post doesn't give a base rate or say how FSD miles are counted. It had 448 likes and 59 replies at pull time.

Check Lemonade here. Farzad's RT here.

"Wrong metric"

In the replies under Farzad's Monday-morning Waymo post, OutofSpecDetail asked at 10:55am CST: "😂 how many cities is Waymo in vs Robotaxi?" Farzad answered at 10:57am CST:

How many cars can Waymo make per week vs Tesla?

Wrong metric <3

FoamingIons pushed back at 11:33am CST: "Production numbers don't matter if the customer ride demand is being fulfilled." Farzad replied at 11:48am CST:

It absolutely does because the higher the production, the higher the supply, the lower the cost for the company with the higher production, the bigger the pricing power, which leads to a competitive advantage Waymo will not be able to overcome.

That's his argument, and "Waymo will not be able to overcome" is his projection. Neither side posts production figures or city counts, and nothing in this pull includes Waymo data.

Between those replies, at 11:24am CST, he posted:

Labor will be a fully autonomous robotic utility in the digital and physical world.

That's the whole post and his projection. It gives no timeline. It had 60 likes and 11,063 impressions. At 1:01pm CST he RTd jamesdouma, from Sunday at 3:54pm CST: "The next 10 years is not going to be an incremental change from the last 10. Ubiquitous robotics plus super intelligence changes everything." That's jamesdouma's view.

Check OutofSpecDetail here. Farzad's "Wrong metric" here. FoamingIons here. Farzad's reply here. Farzad on labor here. jamesdouma here. Farzad's RT here.

"Would love to hear your feedback"

Farzad, Monday at 3:40pm CST:

Would love to hear your feedback on my videos as of late.

What do you like? What do you dislike? What do you miss? What do you wish was there?

It had 175 replies at pull time, and Farzad liked answers in both directions. Joshua Derby (DerbyJoshua) wrote, "I like the simple clear explanations of complex systems," and David Jones (DavidJonesXacct) said "The animations are really helpful." Ulric (eric1021945) called them "Super interesting explainers with amazing detail" but said "I miss the old farzad just talking unscripted." Oh, Sup Y'all (OhSupYall) wrote, "I really truly miss seeing the human Farzad videos of before. I don't watch/read the AI ones." motosixfive said "it feels like watching an infomercial instead of a YouTube video," while calling them "Still good videos." Farzad liked all of those. robot*irl (robotinreallife) quoted the ask with "He's just gonna let the agent read this feedback and make the videos better isn't he? 😂" and he liked that too. This pull has no post from him saying what he'll change.

"Elon's Shocking Optimus Decision" went up Tuesday at 6:04am CST, and a second later he replied "Made with my Grok Bot Solo:" with a link to the bot. At pull time it had 238 likes, 148 bookmarks and 13,601 impressions, the most likes of any video in this pull. He also posted "Why SpaceX Needs 300x More Rocket for AI Satellites" at 5:58am CST Tuesday and "Why Nvidia Cools Its Fastest AI Chips With Liquid Hotter Than a Hot Tub" at 12:12am CST. Since Monday morning he's also posted "Why Blue Origin Is Giving Its Giant Rocket a Third Stage," "Why This Robot Hand Reacts to Touch 4x Faster Than It Sees," "Why Boston Dynamics Ripped Half a Mile of Wire From Stretch," "Why Apple Moved Your iPhone's Memory Off the Chip," "Why This Nuclear Reactor Runs on 220,000 Fuel Balls Instead of Rods," "Why CATL Put Sodium Instead of Lithium in a Car Battery" and "Why NASA Parked Its Mars Probes 1 Million Miles From Earth for a Year." This Digest checks titles only, not the videos' claims, and doesn't quote the narration as Farzad's words.

Check Farzad's ask here. Ulric here. OhSupYall here. motosixfive here. robot*irl here. The Optimus video here. The Solo reply here. Check Why SpaceX Needs 300x More Rocket for AI Satellites here. Check Why Nvidia Cools Its Fastest AI Chips With Liquid Hotter Than a Hot Tub here.