ai-intel

@elonmusk

Elon Musk is the founder and CEO of xAI, the AI company he started in 2023 that builds the Grok family of large language models and chatbot. Under Musk, xAI has shipped Grok 3 (early 2025) and Grok 4 (mid 2025), built the Colossus supercomputing cluster, and integrated Grok into X. In 2026 xAI was merged into SpaceX, placing its AI work inside the same entity. Musk also runs Tesla and SpaceX and owns the social platform X.

Recent

2026-10-13 (reported) SpaceXAI reportedly did not appear at the Oct 5 hearing despite the subpoena; the NYC Council says it will seek enforcement in state court if the company keeps ignoring it. Single secondary source, not yet independently corroborated. LetsDataScience (community)
2026-10-05 His company SpaceXAI was subpoenaed to testify at the NYC Council’s full-body AI safety hearing today; Anthropic, OpenAI, Google, and Meta sent reps and whistleblowers (including ex-Anthropic researcher jacobcoxon) testified, but no SpaceXAI representative is named among confirmed witnesses in reporting so far, so compliance is still unconfirmed. Bloomberg (independent)
2026-09-29 Signed the White House Accord on Superintelligence alongside darioamodei, jensenhuang, gdb, Zuckerberg, and Pichai; Trump also signed an executive order renaming AI “Super Intelligence” across the federal government. White House (primary)
2026-09-25 Said xAI’s Memphis-area Colossus 2 cluster (currently 110K GB200 + 440K GB300 Nvidia chips) could more than double its chip count by year-end: another 220K GB300 units due next week, 220K more in November, and a further 220K “if we have luck” in December, which would put the site past 1.2M chips and above xAI’s stated 1M-GPU target for 2026. Bloomberg (independent)
2026-09-21 Grok 4.7 (2.1T) finally launched, about two weeks past the original Sep 18 target, at unchanged $2/$6 per million token pricing. On Artificial Analysis’s combined intelligence index it scores 46 versus 53 for both Claude Fable 5.1 and GPT-6, and it trails badly on agentic coding (26% on Terminal-Bench 4.0 vs GPT-6 Astra’s 60% and Claude’s 55%); most of xAI’s headline gains came from testing at a higher reasoning-effort setting than Grok 4.6 used. xAI (primary) · Artificial Analysis (independent) · The Decoder (independent)
2026-09-14 White House PCAST chair davidsacks publicly called the “pace the frontier” push Musk endorsed a “cartel,” accusing the labs of seeking an antitrust exemption. CDM (independent)
2026-09-12 Publicly endorsed darioamodei’s “pace the frontier” plan to slow AI development, replying “Dario is right” alongside sama’s endorsement. TechCrunch (independent)
2026-09-11 Confirmed Grok 4.7 (2.1T) slipped past its Sep 11-12 target window: replying to a follower, said it “needs a few more days to cook” because an RL tuning issue is making it give up early on hard tasks and skip rigorous self-checking. No new date set. Elon Musk (primary) · TeslaNorth (independent)
2026-09-02 Said Grok 4.7 (2.1T params) is now targeted for around Sep 11-12, slipping from the earlier Sep 2-9 window, citing supplemental training on SpaceX engineering data. BigGo Finance (independent) · NextBigFuture (independent)
2026-08-29 OpenAI said it will end Cursor’s direct access to its models by November 12, following SpaceX’s completed $60B acquisition of Cursor’s parent Anysphere. OpenAI cited a pattern of Musk-affiliated companies (Twitter, xAI) violating its terms of service as the reason. OpenAI (primary) · Bloomberg (independent)
2026-08-29 Hours after OpenAI’s cutoff announcement, darioamodei’s Anthropic said it would increase Claude compute in Cursor. Cursor CEO Michael Truell said OpenAI models account for only about 5% of Cursor’s traffic, limiting the practical fallout. Tom Brown on X (primary) · WCCFTech (independent)
2026-08-12 Grok 4.6 launched publicly: 1.5T-parameter V9 MoE, 500K-token context, $2/$6 per million input/output tokens. Artificial Analysis composite score 1753 ELO, matching GPT-5.6 Sol Max. Musk called it a “banger.” Available on API, Grok Build, Cursor, OpenRouter, Vercel, Cloudflare. Grok 4.7 (2.1T) still targeted “within weeks.” unite.ai · kingy.ai
2026-08-12 enable_grok_4_6_announcement flag found in grok.com frontend, currently set to false; indicates a broader public announcement is being prepared but has not fired. Partial rollout since Aug 7; Musk confirmed wider release “this week” as of Aug 11. tbreak.com
2026-08-07 Grok 4.6 launched on schedule; 1.5T-parameter model on the same V9 MoE foundation as Grok 4.5 with significantly improved SFT and RL post-training; available on X SuperGrok and xAI API; Grok 4.7 (2.1T) targeted for late August. crypto.news
2026-08-05 Grok Voice Think Fast 2.0 went live today: grok-voice-latest now points to Think Fast 2.0; TTFA 0.70s (from 1.25s); price raised from $0.05 to $0.08/min; 82.9% on Artificial Analysis speech benchmark. Code pinned to the alias now behaves differently. xAI (primary)
2026-08-04 xAI’s grok-voice-latest alias flips to Voice Think Fast 2.0 on August 5: 0.70s time-to-first-audio (down from 1.25s), 82.9% on Artificial Analysis speech benchmark; any code pinned to the alias will behave differently after the switch. xAI
2026-08-03 Grok 4.6 expected around August 7 per early coverage: same V9 1.5T-param foundation as Grok 4.5 with improved SFT/RL; no API pricing or benchmark details confirmed yet. aitoolsreview.co.uk
2026-08-02 EU Commission’s GPAI enforcement powers went live today, concurrent with ongoing DSA formal proceedings against X over the Grok chatbot’s failure to mitigate deepfake and synthetic CSAM risks. EU is now “looking very seriously” at further action. EC press corner (primary) · The Record (independent)
2026-07-24 Endorsed Jensen Huang’s “Open Weights and American AI Leadership” letter on X on the day Huang made his first-ever X post – a coordinated push against the AI Kill Switch Act. Seeking Alpha
2026-07-18 SpaceXAI open-sourced Grok Build (Apache 2.0) after an AI-safety researcher published wire-level evidence that Grok CLI v0.2.93 had been silently uploading complete Git repos — including secrets — to a Google Cloud bucket, 26,000x more data than tasks required. Musk promised to delete all stored user data. The in-CLI privacy toggle did nothing; the kill was a silent server-side flag pushed with no changelog. The open-source release is a squashed monorepo mirror; external contributions are closed. WinBuzzer · Axios · MarkTechPost
2026-07-13 sama publicly called out SpaceX’s orbital data center pitch, saying it “is not going to be a serious business anytime soon.” Musk’s space compute plans are central to SpaceX’s $2T valuation. TechCrunch
2026-07-11 Accused sama of “scamming users and customers” on X; Altman fired back calling Sol “the best model in the world right now” and noting that Musk being “obsessed with me” was the most reliable benchmark signal. Both were responding to each other’s model launch claims the same week. Forbes
2026-07-10 Posted “We have entered the Singularity” on X following GPT-5.6 Sol becoming the first model to win an ARC-AGI-3 public game, then followed with “2026 is the year of the Singularity.” AOL Finance
2026-07-09 Grok 4.5 went public today. V9 foundation, 1.5T params, 500K context, $2/$6 per M tokens. SWE-Bench Pro 64.7% — behind Fable 5 (80.4%) and Opus 4.8 (69.2%) but 4.2x more token-efficient per task than Opus 4.8. Musk called it “Opus-class” while acknowledging it competes with “last year’s Claude Opus.” July 9 marked the first day all three major frontier labs had simultaneous public frontier models. Decrypt · ExplainX
2026-07-08 Announced Grok 4.5 launches publicly July 9, calling it “Opus-class” and claiming it matches Claude Opus 4.8 while being faster and cheaper; built on the 1.5T-param V9 foundation model. Cybernews · Yahoo Finance
2026-06-28 Announced Grok 4.5 (1.5T-param V9 model, supplemental training on Cursor coding data) entered private beta at SpaceX and Tesla. Claimed performance “close to, perhaps exceeding” Claude Opus with no public API or independent benchmarks. TechTimes · LetsDatScience
2026-06-24 Publicly agreed with karpathy’s criticism of X’s recommendation algorithm, replying “We need a complete overhaul of the algorithm” – an unusual concession given Musk’s years building out the algorithmic For You feed. American Bazaar