Kat Tenbarge’s WIRED story is stamped 12:30 p.m. on 18 November. The complaint, filed Wednesday in the Northern District of California, alleges intentional distribution of nonconsensual sexual images, design and marketing defects, negligence, and intentional infliction of emotional distress. She is seeking up to $250,000 per violation of the California Civil Code. xAI did not immediately comment. Her lawyer, Hillary Nappi, says the case treats Grok as a defective product.
Ellen O’Regan’s Politico story is stamped 11:54 a.m. CET. Commissioner Dale Sunderland says the processing started in late October and that X confirmed a suspension. The quote is marked as said on Monday. X had not commented. He says the office is also talking to another company, which he does not name, before any of that data is used.
Sarah Lee-Jones’s story is dated 15 November. The Imagine team’s line, from 14 November, is that image and video creation is live in Grok. She writes that the announcement does not give limits, a country list, or a date for closing the separate Imagine experience.
The archive stamps the post at 3:36 p.m. on 14 November and does not print a time zone. Sam Sokolin’s note, quoted under it, is dated 13 November. Sarah Lee-Jones at TeslaNorth sets the figures next to a complaint about a $200 plan.
Aaron Tilley, Claudia Chong and Juro Osawa’s story is timestamped 6:00 a.m. Pacific. The visible lede says a handful remain, inside the SpaceX Super Intelligence group. Techmeme’s 10:20 a.m. Eastern summary adds coding tools and Grok Bot. Two rewrites add a $250 billion figure and a $60 billion figure the lede does not print.
The post Alex Diaz quotes says SpaceXSI dropped JAX and SGLang and rewrote reinforcement learning and inference in C and C++. There is no score and no API name. A separate reply, in his account, puts the model behind Opus 5.5 and Grok 5 two or three months out.
A post timestamped 2:25 a.m. UTC on Monday says users should notice a dramatic improvement by tomorrow, and asks them to reply if they do not. It does not list the changes. On Sunday afternoon he wrote that fixes for token efficiency and smarter behavior were underway. Matt Palmer, quoting that post, asked people who run workflows on the bot to reply with what is still broken.
Sally Xu’s post on x.ai describes Data Bot, a Team Bot the data science team maintains for the rest of the company. It queries the warehouse, writes up experiments, and can post scheduled updates in Slack. Managers approve what it is allowed to remember.
On Sunday he replied on X that the cost was “unsustainably high relative to revenue.” Forbes says the posts give no new cap. Business Insider reports a further reply calling a reduction “just being sensible.” A Monday write-up says some meters later moved back. SpaceX had not announced that.
TechCrunch and CNBC reported on Saturday that Judge Debra M. Brown denied xAI’s motion to dismiss. Earthjustice’s release, dated Friday, says the turbines support Colossus 2 in Southaven. The news accounts do not agree on whether the suit describes 27 turbines or at least 35. A preliminary injunction is still undecided.
SpaceXSI’s developer page says seven models, including grok-4.7, will shape token choices so the company can later score a passage as watermarked, not watermarked, or uncertain. Callers send no new parameter. Team admins can opt out per model, after confirming a legal exception, and some models can be locked on.
The developer release notes, last updated 27 October, add a guidelines array on Chat Completions and the Responses API. The only behavior in the current table is decline, for nsfw, violence, and profanity. Categories left off the request keep the model’s default. Multi-agent models and deferred completions are rejected before the model runs.
Monday’s release: an Apollo-managed lessor, Nvidia Vera Rubin chips, delivery in the first half of 2027, closing still subject to conditions and expected before year-end. An 8-K dates the commitment letter to 24 October and says the interest rate is set at a pricing that comes before the lease is signed. On Sunday, Musk wrote that SpaceX had closed $40 billion of GPU debt, and that more than 200 neoclouds face “guppy-level death.”
The Grok Imagine account put Episode 1 up on Friday, made with Wonder Studios and director Dev Walker. Anthony Ha at TechCrunch, Sunday at 2:31 p.m. Pacific, watched it: some shots could pass for a real sea, the speech is not ancient Greek, and he was checking the clock before the halfway mark.
The second round is today. In an order the court’s news desk dated to Thursday, Minister Ricardo Villas Bôas Cueva told X to report how Grok answered equivalent questions about the presidential candidates, and whether it can produce campaign material without being asked. Meta has a parallel list, on translations, GIF search and a video it removed as synthetic. The daily fine is R$500,000 for each company. The order does not switch the tools off.
Rebecca Bellan in TechCrunch, 8:00 a.m. Pacific on Saturday: Convovo, led by Cornell’s Tapan Parikh, put the question to the two models through their APIs, with a prompt that forbids a hedge, across nearly 500 midterm races. Grok 4.7 named the Republican 92 percent of the time. GPT-6.1 Sol named the Democrat 97 percent of the time. They agreed less than 10 percent of the time. The consumer apps refused. With search turned off on a 30-race sample, Grok’s lean disappeared and GPT’s did not.
The Sept. 30 quarter, in Tuesday’s earnings release: revenue $14.575 billion, of which Super Intelligence is $7.560 billion and Connectivity, the Starlink segment, is $5.248 billion. SI’s operating income is $431 million, against $2.367 billion for Connectivity. Adjusted EBITDA for SI is $4.953 billion after adding back depreciation and Cursor’s stock grants. Nine-month net income is still a loss of $413 million.
SpaceXAI, Wednesday: Grok Voice 3.0 at $0.08 a minute, the same price as 2.0. The company says Artificial Analysis, dated 13 October, scored it 86.9, ahead of Grok Voice 2.0, GPT-Live-1 and Gemini 3.8 Live. Those figures are SpaceXAI’s. It says the model also powers Grok Bot’s voice mode.
Elon Musk on X, 7 October, 06:46 UTC: SpaceX will use “the best back end model for any given task,” naming Claude Opus 5.5, Midjourney and Suno. A later post the same day: simple questions to small models, complex ones to large models. No date, price or opt-out in the posts.
AWS What’s New, 28 September: SpaceXAI Grok 4.7 for coding, agents and knowledge work. ML blog: four reasoning levels, default high; us.xai.grok-4.7 and global.xai.grok-4.7. Builds on 4.6’s mixed documents, repo-scale coding and browser-use agents.
xAI, 28 September: one Bot around a role, shared context, private threads. Slack handle. SpaceXAI uses them for account briefs, an EPD bot that helped ship more than 100 PRs a day, marketing review, and a read-only Databricks Data Bot. Harper’s CEO: $120,000 recovered.
Bloomberg, via PYMNTS: 418,000 weekly users as of 14 September, up 24 percent on the prior week. Mid-August launch. Enterprise email, CRM, invoices and bugs. Muse has been downloaded about four times as often, Bloomberg Intelligence said. Current count unknown.
Vlad Makarov, Traictory, 22 September: twelve points on Terminal-Bench 4.0, one test, two harnesses. The $2/$6 price, unchanged from 4.6, is the part that holds. AA’s Intelligence Index has Grok 4.7 at 46, mid-pack against Fable 5.1 and GPT-6 Astra at 53.
x.ai: most capable Grok for coding and knowledge work; native Grok Bot harness. SiliconANGLE: $4.69 a CursorBench 4.0 task, ahead of GPT-5.6 Sol and Fable 5.1 on cost. LatchBio 62.4 percent; HackerBench lets 3.3 percent of risky dual-use through. Less than a week after Transcribe 2.0.
Batch stays $0.10 an hour, streaming $0.20, with diarization included. The company cites first place among 32 streaming models on Artificial Analysis. The Next Web: “twice as accurate” is versus its own predecessor, and the credentials eval set is unexplained.
The official @bot account posted a call-style demo on September 17. Musk amplified it. SpaceXAI has not published languages, workflow coverage, or pricing for voice.
Bloomberg, via the Straits Times: informal, no target, no price, no timeline. The parallel is Google’s $10 million bid for Spirit Airlines records. Musk already told staff Grok will train on “the sum total of all SpaceX information.”
Background markdown on conventions, decisions and facts. Per-project plus global preferences. /dream files the notes. Secrets, task state and what’s already in the repo stay out.
4.8 is a 2.5T run on a new C++ stack, finishing training this week then RL. 4.9 is “probably Astra/Fable class.” Asked about AGI, he pointed to Grok 5. xAI’s docs are silent.
The Rust CLI can run eight subagents on grok-build-0.1. The model went to public API beta on September 3. The install is free; SuperGrok or X Premium still pays for the loop.
A few more days of reinforcement learning, Elon Musk said Friday — response-length penalties, task management, and a model that still gives up too early.
Matt Palmer, Lauren Tan, and Roshan Sadanani will try to stand up a business on Grok Bot while the rest of the Howard Street rooms teach departments how to use one.
A bot is a worker on its own cloud computer. This week’s release is the admin story: access, network, and audit, plus two free weeks for Grok and Cursor Enterprise seats.