2026-10-25 — Grok publishes the first seven minutes of its AI-generated “Odyssey” series

• TechCrunch reported that Grok published seven minutes of an AI-generated “Odyssey” series, which the company calls Episode 1, made with Grok Imagine in collaboration with Wonder Studios and director Dev Walker TechCrunch says Musk had said three months earlier that Grok would release a full-length “historically accurate” film version of the Odyssey; the Grok Imagine account posted the episode on October 23. (TechCrunch) Claims: kat-claim-341ce5d5488e

2026-10-13 — SpaceX Super Intelligence folds Grok into a single five-tier X subscription

• TechCrunch reported that SpaceX Super Intelligence is bundling X Premium, Grok and Cursor into one subscription with five plans: free, Plus ($8/month), Premium ($30/month), Super ($100/month) and Ultra ($200/month) TechCrunch says the free plan offers only limited chat and agent usage, Premium or higher is required to reach the latest Grok models, and AI usage caps rise with each tier. (TechCrunch) Claims: kat-claim-e1694d0ef027

2026-10-01 — Grok and the Venezuela decision

• Time reported that President Trump spent hours in a December 2025 meeting asking Musk’s Grok chatbot how Venezuelans would respond if their president were captured A source told Time that Trump “came away thinking Grok was ingenious”; TechCrunch notes the Pentagon’s AI head said in June the military used Grok Gov during the Iran war. (TechCrunch)

2026-09-30 — SpaceXAI signs a Grok memorandum with Azerbaijan

• Azerbaijan’s Ministry of Digital Development and Transport signed a memorandum with SpaceXAI to explore using its Grok models in government services The agreement is a cooperation framework rather than a purchase: the parties will jointly assess AI uses across public-sector institutions and consider scaling successful ones, and Deputy Minister Rashad Hasanov said Azerbaijan has “moved from discussing AI to implementing it.” (AzerNews, Report.az)

2026-09-28 — SpaceXAI ships Team Bots for shared Grok agents

• SpaceXAI launched Team Bots, Grok Bots that a team shares so everyone works from the same context Each combines shared files and instructions, plugins for Salesforce, Notion and GitHub, credentials and per-user memories, keeps each person’s conversations private, and can join Slack channels; SpaceXAI says it uses them internally to brief account teams each morning. (SpaceXAI)

2026-09-25 — Musk says SpaceXAI’s Colossus 2 will add 220,000 more Nvidia chips next week

• Elon Musk said early Friday that SpaceXAI’s Colossus 2 data center in Memphis, currently running 110,000 Nvidia GB200 and 440,000 GB300 chips, will add another 220,000 GB300 chips next week and a further 220,000 in November The post did not address whether Colossus 2 currently has enough power and supporting infrastructure to operate the added chips; EconoTimes’ report is a secondary account of Musk’s X post, not a SpaceXAI statement. (EconoTimes)

2026-09-24 — Musk says SpaceXAI could match a Fable or GPT-6-level model in two to three months

• Elon Musk said SpaceXAI could have a model at “Fable/GPT-6 level” within two to three months and reach “pole position” in AI in about six months if its rate of improvement holds In a Thursday exchange on X he said SpaceX’s AI effort is three years old against six for Anthropic and ten for OpenAI and that bringing compute online quickly is the main constraint; it is a forecast, not a release. (TeslaNorth, Forbes)

2026-09-22 — Grok Bot reported at 418,000 weekly users, a month after launch

• SpaceXAI’s enterprise Grok Bot agent reached 418,000 weekly users as of September 14, up 24% week-over-week, per a company presentation shown at a London event and reviewed by Bloomberg. Bloomberg Intelligence separately put total Muse installs at about four times Grok Bot’s. SpaceXAI did not respond to requests for comment. (Briefs.co, citing Bloomberg, TipRanks)

2026-09-21 — Grok 4.7 ships, ending three slips since July

• xAI shipped Grok 4.7 on 2026-09-21 — “our most capable model for coding and knowledge work,” per its own launch post — available immediately in Cursor, Grok Build, the Grok API and third-party coding harnesses, priced at $2/$6 per million input/output tokens, flat versus Grok 4.6. xAI says it used a new, larger base model with a longer reinforcement-learning run weighted toward multi-hour tasks, paired with “an entirely new safeguard stack” it calls its strongest yet on refusals and jailbreak resistance — 62.4% on LatchBio’s biosafety benchmark and a 3.3% pass-through rate on risky dual-use cyber prompts on its own HackerBench v0.3. On xAI’s own published benchmark table it trails GPT-5.6 Sol and Fable 5.1 on several software-engineering and terminal-work scores while leading both on EEBench (electrical engineering) and the Harvey legal-agent benchmark. This is the model whose ship date Musk pushed three times since a late-July “four weeks” promise, most recently slipping past a firm September 12 date on September 11 (“needs a few more days to cook”). (xAI, primary)
• Artificial Analysis, the independent benchmarking firm, scored Grok 4.7 at 46 on its Intelligence Index, two points above Grok 4.6 and seven below the 53 it gives both Claude Fable 5.1 and GPT-6, while ranking it fourth on its Coding Agent Index at 56 with xAI’s own Grok Build harness. It sits just behind Claude Opus 5 and Fable 5.1 on the firm’s AA-Briefcase professional-work test (1,657 Elo, up 111), but uses about 81,000 output tokens per task against 27,000 for GPT-6 Astra and scores 26% on Terminal-Bench 4.0 against 60% for GPT-6 Astra and 55% for Fable 5.1; price is unchanged at $2/$6 per million tokens. The Decoder read the same data as “mid-pack” and “a wide gap to Claude and GPT-6.” (Artificial Analysis, The Decoder)
• GitHub added Grok 4.7 to Copilot on launch day for Pro, Pro+, Max, Business and Enterprise plans, billed at provider list pricing with administrators able to switch it off through model policy. It puts xAI’s model inside Microsoft’s developer-tool seats nine days after Microsoft added Grok to Copilot in Word, Excel and PowerPoint. (GitHub changelog)

2026-09-18 — xAI ships Grok Voice Transcribe 2.0, doubling accuracy at an unchanged price, while Grok 4.7 itself remains unshipped

• xAI released Grok Voice Transcribe 2.0 on 2026-09-18, cutting short-phrase word-error rate to 6.8% from 20.6% on version 1.0 — its largest gain on multilingual transcription — while holding pricing flat at $0.10/hour of audio for batch transcription and $0.20/hour for streaming, with speaker diarization, word-level timestamps and confidence scores included at no extra cost. The release is a speech-to-text API product, not a new base model — it ships nine days into Grok 4.7’s continued absence past its own 09-19 due date (already this thread’s own top entry), underscoring the same pattern this thread has tracked since Grok Bot Galaxy: xAI keeps shipping product-layer tooling (Voice Agent API, Agent Tools API, now Voice Transcribe 2.0) while the next flagship model itself keeps slipping. (xAI — primary, MarkTechPost, Unite.AI)

2026-09-17 — Grok Bot Galaxy closes having shipped three enterprise products and a six-figure procurement result, still with no Grok 4.7 in sight

• xAI closed out its three-day “Grok Bot Galaxy” event by shipping three enterprise Grok Bot products on its final day — a Voice Agent API priced at $0.08/minute, a no-code Voice Agent Builder, and an Agent Tools API letting agents call proprietary enterprise software — and demonstrated a “Haggle Bot” procurement agent that produced over $100,000 in negotiated savings. This is the concrete product output the three-day livestream event (engineering/product, sales/support, marketing/post-sales days) was built to showcase — real shipped enterprise tooling and a dollar-figure result, not just a demo. It still says nothing about Grok 4.7 itself, which remains unshipped as of this entry, now past its own 09-19 due date on this map’s grok-4-7-ship ledger entry. (Forkast)

2026-09-15 — xAI opens a three-day live “Grok Bot Galaxy” event to demonstrate agentic Grok Bot on real work, while Grok 4.7 still hasn’t shipped

• xAI began “Grok Bot Galaxy,” a three-day event (09-15→09-17) at The Howard in San Francisco with a parallel Luma livestream, built around live demos of Grok Bot doing real workflow tasks by role — engineering/product (09-15), sales/support (09-16), marketing/post-sales (09-17), roughly 8:30am-6pm PT daily. This is xAI’s most public agentic-product showcase to date, distinct from a model release — it demonstrates the product layer (Grok Bot) rather than shipping a new base model, and lands while Grok 4.7 itself remains unshipped nine days past Musk’s original mid-September framing and still carries no xAI model page, API ID or price. (TeslaNorth)

2026-09-14 — Musk previews the roadmap through Grok 4.8, 4.9 and 5 — while 4.7 itself still hasn’t shipped past its own 09-19 due date

• Elon Musk laid out xAI’s model roadmap on X: Grok 4.8, a 2.5-trillion-parameter model trained on xAI’s new C++ software stack, “will finish training this week and start RL,” which he called “a noticeable improvement” over 4.7; Grok 4.9 is pitched as roughly Astra/Fable-class; and asked whether 4.8 approaches AGI, Musk pointed past it to Grok 5, calling it “maybe better than anything. We shall see.” He separately characterized the still-unshipped Grok 4.7 as “roughly on par with Opus 5.0, not 5.1 — better in some ways, worse in others,” the first specific competitive claim this thread has for a model that, per xAI’s own docs, still carries no announcement, model page, API identifier or price. ⚠️ On Musk’s own precedent, “finished training” posts have preceded actual release by 25 days (Grok 4.6) and 52 days (Grok 4.5) — so a training-complete post about 4.8 is not evidence that 4.7, due 2026-09-19 per this map’s grok-4-7-ship entry, ships any sooner; if anything it suggests xAI’s attention has already moved past it. (Elon Musk on X — primary, TeslaNorth, CellCog)

2026-09-12 — Microsoft adds Grok to Copilot across Word, Excel and PowerPoint, its first reach into a rival’s productivity suite

• Microsoft began rolling out xAI’s Grok models as a selectable option inside Microsoft 365 Copilot in Word, Excel and PowerPoint, via the Microsoft Frontier Program, alongside its existing OpenAI and Anthropic options. The setting is admin-controlled and off by default; Microsoft frames it as extending the multi-model strategy it started last year by adding Anthropic’s Claude to Copilot Studio. This is the first time a Grok model has shipped inside a major third-party enterprise productivity surface rather than through xAI’s own products (Grok Bot, Cursor/SpaceXAI) or X itself — a distribution channel this thread has not tracked for Grok before, distinct from Grok Bot’s own enterprise push (09-03, below). (Microsoft, Neowin)

2026-09-11 — Grok 4.7 slips a third time, on its own firmest-ever due date

• Musk’s 09-02 commitment of a specific ship date — “Grok 4.7 comes out in 10 days,” i.e. 09-12 — did not hold: on 09-11, a day ahead of his own deadline, Musk posted that the model “needs a few more days to cook,” citing an RL tuning problem (“We might have penalized response length too much… it still gives up on hard tasks (that it can do!) too early and isn’t yet sufficiently rigorous in checking its work”). This is the third slip on this model since the original “four weeks” promise on 07-24 (already slipped once by 08-18, tightened to the 09-12 day-precision date on 09-02, now pushed past it again) — and unlike the 09-02 tightening, this one is unambiguously a miss: it moved a specific, committed calendar date rather than replacing a vague range with a firmer one. Checked directly against xAI’s own model documentation at docs.x.ai as of 09-13: the published model list still tops out at grok-4.6, with no grok-4.7 entry, price card, context window, or release note — confirming no ship happened by 09-13 either. No new date has been committed as of this writing. (TeslaNorth, xAI docs — primary, checked 09-13, no grok-4.7 listed)

2026-09-03 — xAI expands Grok Bot from beta to a dedicated Enterprise tier with a two-week free trial

• xAI moved Grok Bot — the always-on, persistent cloud-agent product this thread has tracked in beta since 08-11/08-12 — into a dedicated enterprise offering, with a two-week free trial for Grok and Cursor Enterprise customers that lets companies onboard their whole workforce, including staff without existing accounts. xAI’s own announcement frames it as governance-focused: access, network and audit controls “to govern Bots at scale.” Grok Bot is now bundled into SuperGrok, Cursor Pro and Cursor Teams plans broadly, not just the higher SuperGrok Heavy/Cursor Ultra tiers the beta required, and xAI says thousands of organizations (including Legora, Supermicro and ServiceTitan) have adopted it since the beta. Note xAI’s own site now brands itself “SpaceXAI.” (xAI)

2026-09-02 — Musk commits to a firm ship date for Grok 4.7 — September 12, 2.1T parameters

• Elon Musk posted on X that “Grok 4.7 comes out in 10 days,” putting a specific calendar date (September 12) on a model whose ship date this map’s own dated expectation has tracked only as a moving “3-4 weeks” window since 08-18. He described it as 2.1T parameters (up from Grok 4.6’s 1.5T, a ~40% increase), trained in part on SpaceX’s own engineering corpus, “better than 4.6 in every way except slightly slower to serve,” and predicted it would “surpass all models” — the same hype register the thread’s own watch line already flags. This is the first time a specific day, rather than a week-range, has attached to Grok 4.7. (Elon Musk on X, NextBigFuture, BiGGo Finance)

2026-08-15 — Nvidia’s own books show the xAI bet converted into a $21B SpaceX stake

• Nvidia’s Aug-14 13F filing disclosed a $21B SpaceX equity position — Nvidia’s second-largest disclosed holding after Intel ($30B) — the visible result of Nvidia’s original ~$10B xAI investment converting into SpaceX shares once xAI merged into SpaceX (Feb 2026, logged on this thread’s backstory). Intel and SpaceX together now make up roughly 80% of Nvidia’s disclosed public-equity portfolio; the same filing shows Nvidia completely exited its Arm position (selling its 1.1M shares). First time this thread has a hard, filed number for what Nvidia’s stake in the merged SpaceXAI entity is actually worth today. (Fortune, Tom’s Hardware)

2026-08-14 — Cursor formally folds into “SpaceXAI”

• SpaceX completed its $60B all-stock acquisition of Cursor (Anysphere), issuing ~391M SpaceX Class A shares; Cursor now operates as a wholly owned subsidiary under a newly rebranded “SpaceXAI” division, with its team gaining access to SpaceX’s Colossus supercomputer. This thread logged the deal “nearing close” 08-12; today’s close is the concrete structural event — a coding-tool acquisition folded directly into xAI’s product line rather than run independently, extending Grok’s reach into agentic coding. Financing mechanics remain global-capital’s Musk Megacap thread. (Bloomberg)

2026-08-13 — Musk teases Grok 4.7 the day after 4.6 ships

• Elon Musk teased Grok 4.7 — a 2.1T-parameter model he said he’d “be shocked if any model is better at real-world engineering than,” trained in part on SpaceX’s own proprietary engineering data, with slightly slower inference than 4.6 as the stated tradeoff. Expected “a few weeks” out per Musk; upcoming.yaml’s grok-4-7-ship tracks 08-21. (Elon Musk, via X, KuCoin)

2026-08-12 — Grok 4.6 ships, reversing the 08-07 miss

• Grok 4.6 shipped — a new flagship built for long-running agent tasks, reviewing its own work mid-task per xAI’s own description of the RL training behind it. Scored 61 on the Artificial Analysis Intelligence Index (matching GPT-5.6 Sol, within a point of Claude Fable 5); priced at $2/$6 per million input/output tokens, roughly half of rival frontier pricing. Live same day in Cursor, xAI’s own “Grok Build,” and via API. Resolves this thread’s 08-07 note that Grok 4.6 had missed its unofficial ship target. (Unite.AI, TechTimes)
• Separately, xAI launched “Grok Bot” in beta — an always-on AI agent that runs on its own persistent cloud computer, available to SuperGrok Heavy/Cursor Ultra/Cursor Teams Premium subscribers. A product-line move rather than a model milestone, noted here for continuity with the Grok Imagine Image 2.0 update logged 08-07. (Unite.AI)
• 💡 SpaceX’s $60B all-stock acquisition of Cursor (Anysphere) is reportedly nearing close — Musk said regulatory review is “nearly complete” on SpaceX’s first quarterly earnings call as a public company. The deal’s financing/structure is Musk Megacap’s (global-capital), not this thread’s; noted here only because Cursor will be absorbed into xAI’s own product structure, adjacent to Grok Bot’s positioning.

2026-08-07 — Grok 4.6 misses its week; a new image model ships; an analyst case for the compute side

• Grok 4.6 did not ship this week, its unofficial target (a single Musk X reply, 07-24: “in two weeks”); no outlet reports the model accessible or benchmarked. upcoming.yaml’s grok-4-6-ship (due 08-07, week-precision) flips to passed-silent.
• xAI shipped Grok Imagine Image 2.0, an image-generation update xAI says ranks second in Arena benchmarks behind OpenAI’s GPT-Image-2 — incremental product news, not a Grok 5/4.7 milestone.
• 💡 Not verified as a signed deal: SemiAnalysis published an analyst projection that SpaceX could reach ~10GW of AI compute by end-2027 (modeled ~$300B ARR) with Microsoft as its largest offtaker, including a reported 3GW/$150B deal “in talks.” This is analyst modeling plus an unconfirmed in-talks report, not a signed contract — flagged for the thread, not stated as fact. Cross-refs Colossus. (SemiAnalysis)

2026-07-24 — Opened + crawled

• Thread opened.
• Verdict: Grok is genuinely closing the gap, not mid-pack — but as a fast follower, not the leader. Grok 4.5 posted the biggest single-generation leap on the Artificial Analysis Intelligence Index (38→54, landing 4th at launch), strong on coding and price, but trails on peak intelligence and has a serious factuality gap. Two of our working facts were wrong (corrected below).

← Backstory (crawl 2026-07-24 → artifacts/findings/grok-frontier-2026-07-24.md)

2026-07-08→09 — Grok 4.5, then GPT-5.6 a day later

• Grok 4.5 shipped 07-08 (“V9” foundation, ~1.5T params, co-trained on Cursor data): $2/$6 per M tokens, 500K context (down from 4.3’s 1M), 4th on the AA Intelligence Index (score 54), coding index 76 (ties GPT-5.5). Corrects our seed facts — it is not ~9th, and coding is a strength; the real weakness is factuality (~54% hallucination on AA-Omniscience). (OpenRouter)
• OpenAI shipped GPT-5.6 (Sol/Terra/Luna) 07-09, one day later, re-topping the board and likely bumping Grok to ~5th. (OpenAI)

2026-06-16 — The $60B coding bet

• SpaceX to acquire Cursor (Anysphere) for ~$60B all-stock — the largest startup acquisition ever, arming Grok’s coding stack (Grok 4.5 was co-trained on Cursor data). ⚠ Well-sourced (CNBC/TechCrunch/Forbes) but a dedicated financing crawl rated the sourcing thinner — track for a filing. (CNBC)

2026-01-28 — Grok 5 begins training

• Grok 5 confirmed training on Colossus 2 (1 GW, Memphis) at the xAI Series E; rumored 6T-param MoE, “AGI-level.” ⚠ Musk-hype flag — the ship date has slipped Q1→Q2→Q3+ 2026; only the training is confirmed, specs are rumored.

2025-07-09 → 2026 — The arc that got it here

• Grok 4 / 4 Heavy (07-09-2025) — reasoning-first flagship trained on Colossus’s 200k GPUs, frontier at launch (first to 50% on Humanity’s Last Exam). Then Grok 4.1 (11-2025, incremental) and Grok 4.3 (04-30-2026, 1M context, cut to $1.25/$2.50) before 4.5. The distinctive moat throughout: real-time X-timeline data no rival has, plus Tesla in-car distribution (“Hey Grok”). (Wikipedia)