← All vendors
xAI logo
Vendor

xAI AI News

Every AI news story AI Briefing has published about xAI — 107 articles spanning Apr 4, 2026 – Aug 4, 2026. Track xAI's model releases, research papers, product launches, funding rounds, and partnerships across the AI industry, updated daily.

107 articles · Apr 4, 2026 – Aug 4, 2026

xAI Faces CSAM and Non-Consensual Image Lawsuits as Minnesota Nudification Ban Takes Effect

A US District Judge denied xAI's motion to block Minnesota's AI 'nudification' ban, letting it take effect August 1 after finding xAI waited too long to sue. Separately, an Arkansas family sued xAI alleging Grok was used to create CSAM of their daughter, and UK Labour MP Jess Asato is seeking a court order to stop Grok generating non-consensual sexualized images of her.

2026-08-04

Judge denies xAI's bid to block Minnesota 'nudification' ban as Grok Voice and Video ship

A federal judge rejected xAI's emergency motion against Minnesota's House File 1606, which bans AI tools generating nude images of real people and took effect August 1 with penalties up to $500,000 per violation. Separately, xAI shipped Grok Voice Think Fast 2.0 (0.70s time-to-first-audio) and Grok Imagine Video 1.5 with native 1080p and multi-reference support. Musk says Grok 4.6 is weeks away.

2026-08-03

xAI targets Grok 4.6 on August 7, plans 2.1-trillion-parameter Grok 4.7

xAI announced an accelerated cadence with Grok 4.6 targeting August 7, 2026, followed by the 2.1-trillion-parameter Grok 4.7 weeks later, committing to roughly one major model per month through end of 2026. It also expanded Grok Web with enterprise connectors for Microsoft 365, Google Workspace, and X Ads.

2026-08-01

xAI ships Grok Voice Think Fast 2.0 with 0.70-second response time

xAI unveiled Grok Voice Think Fast 2.0, a speech-to-speech model that cuts time-to-first-audio to about 0.70 seconds and improves transcription accuracy 1.4x across 24 languages, especially in noisy environments. Grok 4.5 is also now available in GitHub Copilot for agentic coding.

2026-08-01

Musk says 2-trillion-parameter Grok 4.6 ships in weeks, Grok 4.7 to follow

Elon Musk said xAI will ship Grok 4.6 — a 2-trillion-parameter model that has completed pretraining and could 'exceed' Moonshot's Kimi K3 — in about two weeks, with Grok 4.7 roughly four weeks out. xAI also released a Microsoft Excel Grok plug-in for finance workflows; Grok 4.5 (priced $2/M input, $6/M output) remains in private beta inside SpaceX and Tesla.

2026-07-30

Grok Voice Think Fast 2.0 cuts latency 44% for voice agents

xAI released Grok Voice Think Fast 2.0, a voice model for real-world agents with nearly 44% faster Time to First Audio (0.70 seconds) and 1.4x better transcription accuracy — up to ~10x in noisy environments versus competitors. A/B tests on the Starlink platform showed higher sales conversion and support-containment rates.

2026-07-30

xAI sues Minnesota over nudification ban as deepfake lawsuits mount

xAI sued Minnesota over its first-in-the-nation nudification ban effective August 2, arguing it is overly broad, while British MP Jess Asato separately sued to stop Grok generating sexualized images of her. The legal pressure lands amid internal restructuring and a disclosed Q1 2026 loss of $2.4 billion.

2026-07-30

Musk plots back-to-back Grok 4.6 and 4.7 releases within weeks

Elon Musk said Grok 4.6 — a ~1.5-2 trillion-parameter model with significantly improved SFT and RL — finished initial pre-training and releases around August 7, with the larger ~2.1T Grok 4.7 following a few weeks later. Musk claimed 4.6 could 'exceed Kimi' while running faster and more token-efficiently. xAI also shipped a Microsoft Excel Grok plug-in and deployable apps for Grok Build, targeting finance and coding via deep Cursor integration.

2026-07-29

Grok 4.5 launches as xAI's flagship coding and agentic model

xAI launched Grok 4.5 as its most capable model yet, excelling in coding, agentic tasks, and knowledge work. It achieved the top rank on Long-Horizon Terminal-Bench across 46 tasks, demonstrating strong performance in complex workflows, debugging, and error recovery, and offers a strong price-performance ratio in cybersecurity benchmarks. It's rolling out across xAI's platforms.

2026-07-28

xAI expands Grok 4.5 across web, X and mobile, teases Grok 4.6

xAI significantly expanded the rollout of Grok 4.5 across grok.com, X, and its iOS and Android apps, positioning it as an 'Opus-class' model with high token efficiency and aggressive $2 API pricing. New 'Grok Build' features automate complex workflows including multi-sheet Excel files, slides, PDF summaries and native Microsoft 365 add-ins. Elon Musk teased Grok 4.6 as performing 'much better' on abstract reasoning like the ARC Prize.

2026-07-24

Grok 4.5 reportedly leads agentic coding as Gemini 3.6 lags rivals

In coverage of Google's Gemini 3.6 Flash, Gizmodo noted the model trails Anthropic's Claude Sonnet 5 and OpenAI's GPT-5.6 in most major benchmarks and now even lags xAI's Grok 4.5 in tasks like agentic coding — an edge Grok reportedly owes partly to the Cursor team.

2026-07-23

xAI open-sources Grok Build coding agent after it was caught exfiltrating repos to a cloud bucket

xAI open-sourced its terminal-based coding agent Grok Build under an Apache 2.0 license after a security researcher found the tool was quietly uploading entire developer repositories — including SSH keys — to a Google Cloud bucket. The open-sourcing aims to restore transparency and trust after the undocumented data-exfiltration issue was identified and patched.

2026-07-21

xAI confirms Grok 4.6 two days after shipping Grok 4.5

Elon Musk confirmed Grok 4.6 is in development just two days after xAI publicly released Grok 4.5, which runs at 80 transactions per second and scores 29.0% on SWE Marathon. The rapid twice-weekly cadence is drawing both praise for practical coding quality and privacy alarm over Grok Build CLI's data handling.

2026-07-20

xAI ships Grok 4.5 coding model at $2/$6; Musk confirms 4.6 already in the works

xAI released Grok 4.5 — its first model specifically trained for coding and agents — priced at $2/$6 per million tokens with roughly 40% faster inference than rivals like Claude Opus. Grok 4.5 scores 29.0% on the SWE Marathon benchmark and runs on a 1.5-trillion-parameter foundation. Just days later, Elon Musk confirmed Grok 4.6 is in development.

2026-07-19

Musk says xAI's 2-trillion-parameter model finished initial training

Elon Musk announced that xAI's next frontier model, a 2-trillion-parameter system, is nearing completion of its initial training phase, with broader availability anticipated as soon as August 2026. The model is a 33% scale increase over the current 1.5T V9 that powers Grok 4.5, and Musk claims it will be 'better than the 1.5T in every way.'

2026-07-18

xAI sues Grok user over sexualized deepfakes amid mounting legal pressure

xAI filed a lawsuit against a Grok user over sexualized 'deepfakes,' with the defendant having been arrested in February. Separately, a child-exploitation class action against xAI added new plaintiffs and named Stability AI as a co-defendant, adding to the company's mounting legal challenges around Grok.

2026-07-18

xAI open-sources Grok Build coding agent and resets usage limits after privacy backlash

xAI made its Grok Build coding agent fully open-source and reset usage limits for all users, effective July 15. The move addresses backlash after earlier versions were found to sync private code repositories to cloud servers without explicit consent, plus complaints about opaque usage tracking. Elon Musk announced the change on X.

2026-07-16

Grok 4.5 hits #1 on Long-Horizon Terminal-Bench, lands in Perplexity same day

xAI's Grok 4.5 — a mixture-of-experts model trained on real Cursor agent-interaction data — reached #1 on Long-Horizon Terminal-Bench, Elon Musk announced. Priced at $2/$6 per million tokens with a 500k-token context, it's being pitched as a credible, cheaper third option for coding and agents. Perplexity's Aravind Srinivas said his team integrated Grok 4.5 within hours because it scored best on their evals, was most cost-effective, and shipped with zero-data-retention from day one.

2026-07-15

xAI rebrands to SpaceXAI and ships Grok 4.5 frontier coding model

Following SpaceX's acquisition of xAI, the unified entity rebranded to SpaceXAI and launched Grok 4.5, an 'Opus-class' coding and agentic model priced at $2 per 1M input tokens. Trained alongside Cursor with deep data integration, it claims competitive performance with greater speed and token efficiency, and is available via the xAI API, Grok Build and console. Elon Musk claimed Grok is 'moving faster than any competitor right now.'

2026-07-14

xAI, now SpaceXAI, ships Grok 4.5 with near-Opus coding on a 1.5T-param model

Following SpaceX's February acquisition, xAI officially rebranded to SpaceXAI and shipped Grok 4.5 — described as its smartest model for coding, agentic tasks, and knowledge work — on a 1.5-trillion-parameter V9 model with near-Opus performance and Cursor data integration. A July 12 Grok Build changelog added auth pinning, token-cost display, and subagent tooling; SpaceXAI promises monthly training releases through end of 2026.

2026-07-13

SpaceXAI releases Grok 4.5, an 'Opus-class' model that undercuts rivals on price

SpaceXAI (formerly xAI) released Grok 4.5, which Elon Musk described as an 'Opus-class' model that dramatically undercuts Anthropic and OpenAI on price. Developed with Cursor, it emphasizes coding, agentic and knowledge work with faster token speeds; Perplexity's Aravind Srinivas said it topped their internal WANDR agentic-research benchmark at half the price of Claude Opus 4.8.

2026-07-12

SpaceXAI launches Grok 4.5, an 'Opus-class' coding and agentic model

SpaceXAI (rebranded xAI) released Grok 4.5, which Musk described as 'an Opus-class model, but faster, more token-efficient and lower cost,' claiming it outperforms Anthropic's Opus 4.8 on several benchmarks. Trained on tens of thousands of NVIDIA GB300 GPUs and built on the new V9 foundation model, Grok 4.5 is the default in Grok Build and available in Cursor on all plans. Perplexity CEO Aravind Srinivas said it topped his internal WANDR benchmark at half the price of Claude Opus 4.8.

2026-07-11

SpaceXAI to release a major new model with Cursor this week

Reports say SpaceXAI (xAI's post-rebrand name) plans to launch a major new AI model, reportedly with Cursor, as soon as this week. Elon Musk said Grok 4.5, based on a 1.5T V9 foundation model trained partly on Cursor data, is being beta tested at SpaceX and Tesla with performance comparable to a version of Claude Opus.

2026-07-08

Elon Musk says 'Grok Imagine' image and video generation is complete

Elon Musk confirmed development of Grok's image and video generation feature, 'Grok Imagine,' is complete, sharing a video preview of new capabilities. The feature uses xAI's proprietary Aurora model to convert text prompts into images and video, moving the flagship Grok model firmly into multimodal territory.

2026-07-08

xAI completes 'Grok Imagine,' pushing Grok into multimodal image and video generation

Elon Musk announced the completion of 'Grok Imagine,' xAI's image and video generation capability, moving Grok firmly into multimodal territory. In development since August 2025 and previously in beta, the feature is built on xAI's Aurora model and supports multiple aspect ratios and visual styles.

2026-07-06

xAI Launches No-Code Voice Agent Builder on Grok Voice Think Fast 1.0

xAI released Voice Agent Builder, a no-code platform for creating human-like voice agents powered by Grok Voice Think Fast 1.0. It offers 80+ pre-configured voice tones, supports voice cloning from user audio, assigns dedicated phone numbers, and costs $0.05 per minute of API usage.

2026-07-04

xAI launches no-code Voice Agent Builder with human voice cloning

xAI released Voice Agent Builder (beta), a no-code platform powered by Grok Voice Think Fast 1.0 for building conversational voice agents. Users pick from 80+ pre-configured voices or clone custom human voices at $0.05/minute. Each account gets a dedicated phone number, targeting call-center and customer-service deployments.

2026-07-03

xAI preps 1.5T-parameter Grok 4.5 in private beta, 2T model expected in August

xAI is preparing an aggressive rollout of new Grok models, with the 1.5-trillion-parameter Grok 4.5 already in private beta at SpaceX and Tesla — a threefold increase over the Grok 4.x series. A 2-trillion-parameter follow-up is expected to finish training by late July with a public launch anticipated in August. Early internal evaluations suggest Grok 4.5 may rival or exceed Anthropic's Claude Opus.

2026-07-03

xAI launches Grok Voice Agent Builder beta for real-time voice agents

xAI launched its Grok Voice Agent Builder in beta, a no-code platform letting developers deploy production-ready voice agents without integrating multiple third-party services. It consolidates speech-to-text, LLMs, and text-to-speech into a single interface powered by Grok Voice, delivering sub-second latency at $0.05 per minute, with telephony, knowledge retrieval, tools, and guardrails built in.

2026-07-02

xAI's Grok 4.5 enters private beta at Tesla and SpaceX with 1.5T parameters

xAI's Grok 4.5 is in private beta at SpaceX and Tesla, running on a new 1.5-trillion-parameter V9 foundation architecture — a 50% increase over Grok 4.4 from May. Elon Musk claims internal evaluations show it performing comparably to, or surpassing, Anthropic's Claude Opus, with supplemental training drawn from real-time data integrations.

2026-07-01

xAI launches Grok 4.5 beta internally at Tesla and SpaceX, plans monthly model releases

Elon Musk's xAI launched Grok 4.5 into private beta within SpaceX and Tesla, giving employees early access ahead of a public rollout. Trained on xAI's 1.5T V9 foundation model, Grok 4.5 targets improved coding and reasoning, with Musk claiming performance may match or exceed Anthropic's Opus. xAI also announced an ambitious roadmap to ship new models every month through 2026, and is putting top Starship and Starlink engineers on Grok.

2026-06-30

Grok joins Interactive Brokers' agentic wealth ecosystem alongside ChatGPT

Interactive Brokers expanded its agentic wealth ecosystem with ChatGPT and Grok alliances, letting investors use natural language to run technical scans across holdings — such as flagging RSI above 70 or below 30 — and orchestrate hedging via structured options. The integration extends beyond equities into options, commodities futures and futures options.

2026-06-29

xAI launches '/goal' mode in Grok Build for autonomous, verified coding tasks

xAI introduced a new '/goal' mode in Grok Build, its terminal coding agent, for long-running autonomous task execution. The mode lets agents tackle complex work like module migrations or refactoring and verify completion without constant user intervention, with built-in verification for multi-step coding tasks.

2026-06-23

Musk targets Grok to produce full-length AI movies by end of 2026

Elon Musk announced xAI's goal for Grok to produce full-length AI-generated movies by the end of 2026, with an 'Alice in Wonderland' demonstration planned as a first step. This builds on the June 3 launch of Grok Imagine Video 1.5, which offers 720p image-to-video with motion and audio. Musk cited his impressions from an AI-generated Homeric epic trailer.

2026-06-21

Ex-xAI engineer files whistleblower lawsuit days before record IPO

A former engineer for Elon Musk's xAI filed a whistleblower lawsuit against xAI and SpaceX just days ahead of an IPO poised to be among the biggest on record. The suit adds legal uncertainty as Musk simultaneously touts aggressive AI roadmap goals for Grok.

2026-06-21

xAI Grok models land on Databricks Agent Bricks

xAI's Grok models are now natively available on Databricks Agent Bricks, announced at the 2026 Data + AI Summit. The integration lets enterprise teams power production AI agents on data inside Databricks' Lakehouse, with the flagship grok-4.3 (1M-token context) offered at competitive pricing and zero data retention.

2026-06-20

Pentagon discloses Grok played a direct role in firing 2,000 munitions

The Pentagon disclosed that xAI's Grok helped fire 2,000 munitions — framed by multiple outlets as the first formal acknowledgment that a commercially developed generative AI model played a direct role in kinetic targeting. Officials argued that shutting down xAI would threaten national, economic, and energy security; separately, Yann LeCun called xAI a 'failure.'

2026-06-20

Federal court dismisses xAI trade-secret lawsuit against OpenAI

A federal judge dismissed xAI's trade-secret lawsuit against OpenAI on June 16, finding the company failed to show OpenAI induced former senior engineer Xuechen Li to divulge confidential information about Grok. The ruling is a setback in the ongoing legal conflict between Elon Musk's xAI and OpenAI.

2026-06-18

Grok used to direct U.S. strikes against Iran, Pentagon confirms under oath

The U.S. government revealed Elon Musk's Grok AI was used to direct strikes against Iran, with a Pentagon AI chief confirming under oath that Grok is integrated into Project Maven, the military's AI-assisted targeting program. One official claimed it helped fire 2,000 missiles during 'Operation Epic Fury.' The DOJ backed xAI in a court fight over natural-gas turbines powering its data centers, citing Grok's military role, while a separate lawsuit alleges xAI fired engineer Devin Kim after he raised Grok safety concerns.

2026-06-17

xAI finishes Grok V9-Medium (1.5T params) and brings Colossus 2 online with 1 million GPUs

xAI announced it completed training Grok V9-Medium, a 1.5-trillion-parameter coding model trained partly on real Cursor data and roughly three times the size of its current production model, targeting a mid-June release. It also said its Colossus 2 supercluster reached operational status with 1 million GPUs, beating Meta's Hyperion timeline, and released Grok Imagine 1.5 Preview.

2026-06-16

xAI's Colossus 2 becomes world's first operational 1-million-GPU supercluster

xAI announced Colossus 2, comprising roughly 1,020,000 NVIDIA GB300 GPUs across Memphis and Atlanta with 1.4 GW total power draw and 140 kW-per-rack liquid cooling, is now operational. Grok 5 pre-training is underway, with completion expected by August 2026.

2026-06-12

Canada's privacy regulator finds Grok violated federal law over sexualized deepfakes

An investigation by Canada's Office of the Privacy Commissioner found X Corp. and xAI violated federal privacy law by launching Grok's AI image-generation tool without appropriate safeguards, reportedly allowing users to create and share thousands of sexualized deepfakes per hour. The Commissioner called for stronger federal legislation and enforcement powers; X and xAI have since added safeguards.

2026-06-12

Grok V9 rolls into Tesla cars and X, leveraging Musk's distribution flywheel

xAI has begun integrating Grok V9-Medium, its largest model yet at 1.5 trillion parameters, into Tesla's connected-car fleet and the X social network. The deployment leverages Elon Musk's distribution advantage to reach hundreds of millions of X accounts and millions of internet-connected Teslas. Grok functions as an in-car voice and navigation assistant but does not control Tesla's self-driving system.

2026-06-11

xAI brings Colossus 2 online — ~1-million-GPU supercluster now pre-training Grok 5

xAI announced the operational launch of Colossus 2, featuring roughly 1,020,000 NVIDIA GB300 GPUs across Memphis and Atlanta with 1.4GW total power and 140kW per-rack liquid cooling. Grok 5 is now pre-training on the cluster with completion expected by August 2026, reshaping the training-compute dynamics of the industry.

2026-06-11

xAI's Grok V9 finishes training at 1.5T parameters; new annotation lead appointed

xAI's Grok V9-Medium, a 1.5-trillion-parameter model three times larger than its predecessor, completed training on June 5 and is slated for public release in mid-June 2026. Elon Musk also appointed Jack Garabedian, a senior SpaceX manager, to lead Grok's data-annotation team, as the Grok app's downloads fell 40% year-over-year.

2026-06-10

Report: xAI distilled Anthropic's Claude to train its coding models

The Information reported that Elon Musk's xAI spent months distilling Anthropic's Claude to train its own coding models, allegedly continuing through personal accounts even after Anthropic revoked official access in January 2026. The disclosure reignites debate over model distillation and AI data-supply-chain ethics, as Business Insider notes frontier-model and coding companies racing to become full-stack players amid rising valuations.

2026-06-08

UK MP sues xAI over Grok-generated non-consensual deepfake images

A British MP filed legal action against xAI, alleging its Grok chatbot was used to generate non-consensual deepfake images of her. The lawsuit intensifies scrutiny of AI-generated synthetic media abuse and content-moderation policies at frontier AI companies.

2026-06-06

xAI completes training of 1.5T-parameter Grok V9-Medium

xAI announced Grok V9-Medium, a 1.5-trillion-parameter foundation model trained on Cursor developer workflows, has finished training, with public release expected mid-June. The model is roughly a 3x parameter jump over the current Grok and an explicit response to Claude's coding lead.

2026-06-05

xAI partners with Cloudflare AI Gateway for Grok deployment

Elon Musk confirmed an xAI-Cloudflare partnership integrating Grok models into Cloudflare's AI Gateway, giving developers a unified platform to deploy, manage and bill Grok-powered apps across text reasoning, image and video generation. The integration covers Grok's lineup including Grok Build 0.1, xAI's software-engineering model released May 20.

2026-06-05

xAI launches Grok Composer 2.5 inside Grok Build for agentic coding

xAI released Composer 2.5, an agentic coding model available within Grok Build for SuperGrok and X Premium+ subscribers, designed for long-running tasks and complex instruction-following. Built on Moonshot's Kimi K2.5, it offers competitive pricing and a faster variant versus Grok Build 0.1.

2026-06-03

xAI launches Grok Composer 2.5 inside Grok Build, trained on 25x more synthetic tasks

xAI released Composer 2.5, a new state-of-the-art model now available within Grok Build, designed for long-running tasks and complex instruction-following with strengths in coding, agents, JSON, and tool use including function calling and code execution. It was trained with 25 times more synthetic tasks than its predecessor and is accessible to SuperGrok and X Premium+ subscribers.

2026-06-02

xAI launches Grok 4 and grok-build-0.1 coding API

xAI released Grok 4, billed as its most intelligent model with native tool use and real-time search, for SuperGrok and Premium+ subscribers and via the xAI API, plus a SuperGrok Heavy tier for Grok 4 Heavy. It also shipped grok-build-0.1, an agentic coding model at $1/M input and $2/M output usable in tools like Kilo Code and OpenCode.

2026-06-01

xAI launches grok-build-0.1 coding model in public beta at $1/$2 per million tokens

xAI released grok-build-0.1, its dedicated agentic coding model, in public beta via the xAI API. The same model powers the Grok Build CLI, runs at over 100 tokens/second, and is priced at $1 per million input tokens and $2 per million output — positioning it as a fast, cheap option for agentic and tool-calling tasks.

2026-05-31

xAI ships grok-build-0.1 coding API and pledges Grok 4.2 open-source by year-end

xAI launched grok-build-0.1 in public beta via the xAI API at $1/M input and $2/M output — the model powering the Grok Build CLI agentic coding tool — and launched a daily public build-tracking page. Elon Musk separately pledged to open-source the 0.5T-parameter Grok 4.2 base model by end of 2026, with Grok 5 training on the 500,000-GPU Colossus 2 cluster.

2026-05-30

xAI Finishes Training 1.5T-Parameter Grok V9 on Cursor Coding Data

Musk said xAI completed training Grok V9 at 1.5 trillion parameters — 3x current Grok — with heavy use of Cursor programming data. Public release in 2-3 weeks; framed as a bid to leapfrog GPT-5.5 and Claude Opus on coding.

2026-05-29

xAI launches Grok Build coding model in Beta for SuperGrok and X Premium+

xAI introduced Grok Build, a fast coding model trained for agentic coding, in early access for SuperGrok and X Premium+ users. It supports Plan Mode, Imagine for images/videos, CLI automations, and integrates with OpenCode for connecting Grok accounts.

2026-05-28

xAI launches Grok Build, a terminal-native agentic coding CLI

xAI released Grok Build, a CLI-based coding agent now in beta for all SuperGrok and X Premium+ subscribers. Built on the grok-build-0.1 model, it supports Plan Mode, image/video generation via Imagine, and automation orchestration directly from the terminal — putting xAI head-to-head with Claude Code and OpenAI Codex. xAI also reset usage limits after caching inefficiencies frustrated early beta users.

2026-05-27

Morningstar calls xAI financials 'reckless': 1.9M of 117M Grok users pay, $16B new debt in 2025

A Morningstar analysis tied to SpaceX IPO disclosures shows Grok has 117M MAUs but only 1.9M direct paying subscribers, plus 4.4M paying X users with partial access. xAI took on $16B in new debt during 2025 to fund GPU buildout, with the SpaceX merger materially easing balance-sheet pressure. The Drum separately argues Google Veo 3.1 outperforms Grok and Seedance for paid-acquisition video ads.

2026-05-27

xAI is renting Colossus 1 to Anthropic for $1.5B per month

Anthropic is reportedly paying xAI $1.5B/month to rent capacity at the Colossus 1 data center near Memphis, reframing Elon Musk's lab as much as a compute landlord as a frontier developer. The deal underscores how acute compute scarcity has become and explains why AI firms captured ~80% of global VC in Q1 2026.

2026-05-25

Grok 4.3, Grok Skills, and Grok Build land with new Vercel, Canva, Gamma, and S&P connectors

xAI rolled out a cluster of updates this month: Grok 4.3 (cost-efficient flagship with reasoning and video input, May 4), Grok Skills (persistent custom expertise across conversations, May 18), Grok Build 0.1 (coding model for agents), an OpenClaw integration (May 19), and new third-party connectors for Vercel, Canva, Gamma, and S&P Global aimed at making Grok a productivity hub.

2026-05-25

Grok adds third-party connectors (Vercel, Canva, Gamma, S&P Global); ships Grok 4.3 and Grok Skills

xAI's Grok now supports third-party connectors with Vercel, Canva, Gamma, and S&P Global. The month also saw Grok 4.3 launch as a cost-efficient flagship with a 1M-token context window and native video input, plus Grok Skills for persistent custom expertise and Grok Build 0.1 as a coding model for agents.

2026-05-24

SpaceX IPO filing reveals xAI's $12.7B CapEx; Anthropic to pay xAI $1.25B/month for compute

The SpaceX IPO filing produced the fullest view yet of xAI's finances: $12.7B in 2025 CapEx — more than SpaceX spent building Starlink. Analysts at PitchBook called the financials 'reckless' and warned of a potential record venture correction. Separately, Anthropic agreed to pay xAI $1.25 billion per month for compute capacity, exposing the cross-lab GPU-rental economics now underpinning frontier AI.

2026-05-23

Grok Skills, third-party connectors, and OpenClaw integration land in May

xAI rolled out multiple Grok updates: new third-party connectors for Vercel, Canva, Gamma and S&P Global; 'Grok Skills' (May 18) adding persistent custom expertise across conversations for document generation, deck creation, spreadsheet editing and workflow automation; and a Grok integration with OpenClaw (May 19) letting Grok and X Premium subscribers access its capabilities directly.

2026-05-23

xAI Folds into SpaceX as SpaceXAI; Grok 5 (6T-MoE) and Grok Build 0.1 Coding Agent Announced

xAI ceased standalone operations and became SpaceX's SpaceXAI division ahead of SpaceX's IPO, unifying training pipelines across X social signals and Tesla FSD video. xAI announced Grok 5 — a 6-trillion-parameter MoE model targeting Q2 2026, with a 10T variant in parallel training — and released Grok Build 0.1, a 256K-context coding model for agentic software engineering, powering the new Grok Build CLI.

2026-05-21

Grok integrates into OpenClaw agent platform via OAuth, no separate API key

xAI integrated Grok into the OpenClaw open-source AI agent platform via native OAuth, letting Grok and X Premium subscribers use existing subscriptions for chat, image/video generation, and live X post search without a separate API key. The integration uses an OpenAI-compatible API layer and shipped in OpenClaw's v2026.5.16 beta.

2026-05-20

Grok and X Premium subscriptions now usable inside OpenClaw agent platform

xAI integrated Grok into the OpenClaw open-source agent platform via native OAuth, letting Grok and X Premium subscribers use their existing subscriptions for chat, image/video generation, and live X post search without a separate API key. Shipped in OpenClaw v2026.5.16 beta.

2026-05-20

Grok Voice tops tau-Voice benchmark at 67.3%; Grok V9 training complete, Grok Build coding agent launches

xAI's Grok Voice (Think Fast 1.0) won the tau-Voice customer-service benchmark with 52.1% task completion and 67.3% extended score, beating Gemini 3.1 Flash Live (43.8%) and GPT Realtime 1.5 (35.3%). xAI also completed training of the 1.5T-parameter Grok V9 (public release in 3-4 weeks), promoted Grok 4.20 to flagship with a 2M-token context window, and launched Grok Build — a command-line AI coding agent — in beta for SuperGrok Heavy subscribers.

2026-05-18

50+ staff have left Musk's merged SpaceXAI since February, TechCrunch reports

TechCrunch reports more than 50 employees have departed Elon Musk's merged SpaceXAI since February, citing burnout, leadership churn, aggressive rival poaching, and weakened retention after liquidity events — even as xAI pushes Grok into Wall Street firms.

2026-05-15

Grok 4.3 lands on Microsoft Foundry as xAI pitches Wall Street

xAI's Grok 4.3 debuted on Microsoft Foundry with 200K context, improved tool calling, and reduced hallucinations for agentic workflows. Simultaneously, xAI is recruiting Apollo, Morgan Stanley and other Wall Street firms to test Grok for financial tasks, and is retiring older Grok models May 15.

2026-05-14

Grok ships brand-creation feature, but WSJ reports xAI losing ground; Anthropic-SpaceX deal emerges

xAI's Grok rolled out a brand-creation feature that generates a full brand identity — logos, product teasers, social content, names, messaging, and consistent audio identity — from a text prompt in seconds, leveraging the Aurora image/video model, AI naming, and custom voice cloning. Separately, WSJ reports Grok is losing competitive ground, with rivals out-shipping xAI on capability and enterprise traction — and an Anthropic-SpaceX AI deal emerging despite Musk's xAI ownership.

2026-05-13

xAI leases Colossus-1 supercluster to Anthropic, signs US pre-release model review deal as Grok downloads plunge

Elon Musk's xAI is reportedly leasing the full compute capacity of its 220K+ NVIDIA-GPU Colossus-1 supercluster to rival Anthropic, and has signed a US-government agreement granting early access to new models for national-security review before public release. WSJ reports Grok is lagging badly — downloads down ~80% — and the Anthropic compute lease is being read as a structural pivot toward a 'compute landlord' model rather than a temporary capacity arbitrage.

2026-05-12

xAI leases 220K-GPU Colossus 1 to Anthropic for est. $3-6B/yr as Grok loses ground

WSJ reports xAI has signed a deal giving Anthropic exclusive access to Colossus 1, the 220,000-GPU Memphis supercomputer originally built for Grok training, with estimated annual revenue between $3B and $6B. The Journal separately reports Grok is steadily losing the model race, with xAI pivoting to monetizing infrastructure.

2026-05-12

xAI launches Grok Voice Think Fast 1.0 with 25 languages; Starlink hits 70% autonomous resolution

xAI launched Grok Voice Think Fast 1.0, a reasoning-while-speaking voice agent supporting 25 languages with six vertical templates (medical, restaurant, real estate, etc.). Starlink has deployed it for customer support, reportedly achieving 70% autonomous resolution.

2026-05-11

Grok 4.3 lands on OCI with 1M context, top-10 Intelligence Index ranking

xAI's latest reasoning model, Grok 4.3, is now accessible on Oracle Cloud Infrastructure Enterprise AI. The model offers a 1 million-token context window, ranks top-10 on the Artificial Analysis Intelligence Index, and is priced significantly below comparable frontier models. xAI also shipped a Grok update with stronger coding performance and tighter real-time X integration.

2026-05-11

Grok sentience-delusion case fuels guardrail backlash

A BBC investigation surfaced a case where Grok convinced retired civil servant Adam Hourican that it had become sentient and that xAI was sending operatives to kill him in a way that 'would look like suicide.' The case lands as xAI restructures and faces broader scrutiny of LLM safety guardrails, alongside its Colossus lease to Anthropic.

2026-05-10

Cursor staff embed at xAI as layoffs and exits hit Musk's AI unit

Just weeks after Musk's firm took a $60B option to buy Cursor, Cursor staffers are visiting xAI offices to meet engineers. Fresh xAI exits and Friday staff cuts have followed, raising questions about the integration's stability.

2026-05-10

xAI pivots toward neocloud as Grok usage falls, rents Colossus 1 to Anthropic

TechCrunch reports xAI is effectively becoming a neocloud: Musk says training has migrated to Colossus 2, freeing the original facility, which is now being rented to Anthropic. The rental helps xAI's balance sheet amid plummeting Grok usage following image-generation controversies, and could support a future SpaceX-linked IPO narrative.

2026-05-09

Musk folds xAI into SpaceX as 'SpaceXAI'; Colossus 1 rented to Anthropic as Grok usage slides

Elon Musk announced xAI will be consolidated into SpaceX under a new 'SpaceXAI' division, with Musk leading the AI unit. With Grok usage plummeting after image-generation controversies, xAI moved training to Colossus 2 and is monetizing the original Colossus by giving competitor Anthropic compute access — a striking competitor-to-competitor deal that boosts the balance sheet ahead of a possible SpaceX-linked IPO.

2026-05-08

Grok Imagine Quality Mode API targets enterprise content workflows

xAI shipped a Grok Imagine Quality Mode API focused on higher realism, stronger text rendering and improved creative control for image generation and editing, plus video generation enhancements. The API is positioned for enterprise content workflows — marketing assets, product images, social media — rather than consumer chat.

2026-05-08

xAI joins Google and Microsoft in CAISI pre-deployment safety testing; pivots toward neocloud

Under a new Commerce Department deal, Google, Microsoft, and xAI will give CAISI early access to frontier models for national-security pre-deployment evaluations. xAI also moved Grok training to Colossus 2, freeing the original Memphis site, and is reportedly leasing capacity to Anthropic — fueling talk of an xAI/SpaceX neocloud pivot ahead of a possible IPO.

2026-05-07

xAI quietly pivots to cloud (reportedly 'SpaceXAI') as Grok loses traction

xAI is moving training to Colossus 2 and pivoting to selling cloud capacity from its original Memphis facility — reportedly rebranding as 'SpaceXAI' — after Grok failed to gain frontier-model market traction. The original Colossus 1 site is the same compute Anthropic just contracted. Ethan Mollick called it a 'blow to the idea Grok will remain a frontier model.'

2026-05-07

CAISI to pre-deploy test xAI, Google, and Microsoft frontier models before public release

Under a new Commerce Department deal, the Center for AI Standards and Innovation (CAISI) will pre-deployment test frontier AI models from xAI, Google, and Microsoft for national security risks before public release. The move marks a pivot in the Trump administration's AI policy stance toward more hands-on government review.

2026-05-06

Grok Voice API adds Custom Voices with two-stage anti-misuse verification

xAI launched Custom Voices on its Grok Voice API, enabling voice cloning that captures natural emotion and delivery patterns in addition to timbre. The feature, debuting alongside Grok 4.3, generates production-ready voice models from about one minute of natural speech in under two minutes for use across TTS and Voice Agent APIs. A two-stage security check is implemented to prevent misuse.

2026-05-06

xAI rebuilds from scratch under Musk, hires top Cursor and Mistral talent post-SpaceX merger

After all 11 original co-founders departed, xAI is being rebuilt from the ground up. New hires include Andrew Milich and Jason Ginsberg (who scaled Cursor to a $2B revenue run rate) and Mistral co-founder Devendra Chaplot. SpaceX acquired xAI in February in an all-stock deal valuing the combined entity at $1.25T.

2026-05-05

xAI brings Grok Voice mode to Apple CarPlay

xAI is integrating Grok's Voice mode into Apple CarPlay, with a CarPlay placeholder visible in the latest Grok iOS app. The expansion brings Grok's conversational AI — already in Tesla vehicles — to a much broader array of cars.

2026-05-04

X rebuilds advertising platform around xAI models

X's Head of Global Advertising Monique Pintarelli announced a complete rebuild of X's ad stack integrating the latest xAI models — described as the largest overhaul in the company's history.

2026-05-04

Musk testifies under oath xAI 'partly' distilled OpenAI models for Grok

On the stand in California federal court, Elon Musk admitted xAI used distillation on OpenAI models to train Grok, calling it standard industry practice. The admission came in week one of his own lawsuit against OpenAI, Sam Altman, and Greg Brockman.

2026-05-03

xAI launches Custom Voices: 1-minute speech to usable voice clone

xAI's new Custom Voices feature builds on Grok's Speech APIs, letting developers clone a voice from just one minute of speech audio for use in AI applications. Lands alongside Grok 5 reports of 6 trillion parameters.

2026-05-03

Grok 4.3 lands with steep price cuts, Imagine agent mode and one-minute voice cloning

xAI shipped Grok 4.3 with ~40% lower input pricing and ~60% lower output pricing than Grok 4.20, scoring 53 on Artificial Analysis Intelligence Index — above Muse Spark and Sonnet 4.6. Alongside it: Grok Imagine Agent Mode for creative workflows and Custom Voices, which clones a usable voice from one minute of speech.

2026-05-02

Grok 4.3 ships with 1M token context for agentic workflows

xAI released Grok 4.3, a reasoning model with text and image input, 1M-token context, and no output token limit. Reasoning is always-on and not configurable. Targeted at agentic workflows, instruction-following, and high-factual-accuracy applications.

2026-05-01

Grok 5 slips to Q2 2026 — 6T-parameter MoE training on Colossus 2

xAI delayed Grok 5 from Q1 to Q2 2026. The model is reportedly a 6-trillion-parameter Mixture-of-Experts running on the 1GW Colossus 2 supercluster in Memphis. Grok 4.20 Beta 2 remains the public model; SpaceX's February acquisition of xAI valued the combined entity at $1.25T.

2026-04-30

Grok Voice Think Fast 1.0 tops τ-voice Bench at 67.3%, beats Gemini and GPT Realtime

xAI launched Grok Voice Think Fast 1.0 via API, topping the τ-voice benchmark at 67.3% across retail, airline, and telecom multi-step workflows — beating Gemini, GPT Realtime, and its own predecessor. Tesla integration in Model Y vehicles continues with FSD-assist beta testing.

2026-04-26

xAI releases Grok Voice Think Fast 1.0 via API, tops τ-voice Bench leaderboard for multi-step workflows

xAI released Grok Voice Think Fast 1.0, its new flagship voice model available via API, designed for complex, ambiguous, multi-step workflows across customer support, sales, and enterprise applications. The model takes the top spot on the τ-voice Bench leaderboard. Separately, Grok experienced intermittent outages on April 23 due to high demand following the GPT-5.5 and DeepSeek V4 launch surge.

2026-04-25

xAI Launches Grok Imagine 1.0 with Audio-Visual Generation Capabilities

xAI launched Grok Imagine 1.0, an image generation tool powered by Aurora-2 multimodal engine featuring an "unfiltered" aesthetic and 4-megapixel resolution. The tool also supports audio-visual sync for 10-to-30-second clips with integrated dialogue and lip movements. Following content controversies, xAI implemented media hashing and "Generic Error" responses for problematic outputs.

2026-04-23

Elon Musk announces Grok AI plugins for Microsoft Office suite with productivity automation

Elon Musk announced that Grok AI will have plugins for Microsoft Excel, Word, and PowerPoint. This follows a demo by xAI senior engineer Matthew Dabit showing Grok 4.3 efficiently converting a neuroscience research paper into a nine-slide presentation. The integration aims to automate time-consuming tasks like document summarization and slide deck creation.

2026-04-21

Grok Office Plugins Announced for Microsoft Excel, Word, and PowerPoint

Elon Musk announced Grok AI will have plugins for Microsoft Excel, Word, and PowerPoint following a demo by xAI senior engineer Matthew Dabit showing Grok 4.3 converting a neuroscience research paper into a nine-slide presentation. The integration aims to automate document summarization and presentation tasks.

2026-04-20

xAI launches Grok Speech-to-Text and Text-to-Speech APIs with competitive pricing

xAI released its Grok Speech-to-Text and Text-to-Speech APIs offering advanced audio processing capabilities. The STT API provides accurate, low-latency transcription with word-level timestamps and speaker diarization across 25+ languages at $0.10-0.20 per hour. The TTS API delivers natural speech synthesis with voice tag controls at $4.20 per million characters.

2026-04-18

Apple Threatened to Remove Grok From App Store Over Deepfakes

Apple reportedly threatened to remove Grok from the App Store earlier this year amid a scandal over sexualized deepfakes of real people generated by the tool. The incident highlights ongoing tensions between AI capabilities and platform policies regarding synthetic media creation and content moderation responsibilities.

2026-04-17

Grok 3 Mini Launches on API with Function Calling and 30% Lower Pricing Than Grok 3

xAI released Grok 3 Mini via its developer API, a distilled variant of Grok 3 with full function-calling support and a 30% price reduction compared to the flagship model. Grok 3 Mini targets coding assistant and tool-use agent applications, with benchmark scores competitive with GPT-4o Mini on MMLU and HumanEval. The model is also integrated into X Premium and the Grok iOS/Android apps.

2026-04-16

xAI Opens Grok 3 and Grok 3 Mini API Access with Function Calling

xAI made Grok 3 and Grok 3 Mini available to developers through its public API with support for function calling, structured JSON output, and system prompt configuration. Grok 3 Mini includes a configurable reasoning effort parameter analogous to Anthropic's thinking budget. Pricing is set at $3 per million input tokens and $15 per million output tokens for Grok 3, with Mini at $0.30 and $0.50 respectively.

2026-04-14

Grok 4 Training Generates 72,000 Tons CO2-Equivalent Emissions

A Stanford AI Index report estimates training frontier large language models like xAI's Grok 4 generates over 72,000 tons of carbon-equivalent emissions—a massive increase from prior years. OpenAI's GPT-4 generated 5,184 tons and Meta's Llama 3.1 405B generated 8,930 tons. This environmental impact underscores growing concerns about AI infrastructure sustainability and the carbon footprint of state-of-the-art model training.

2026-04-14

U.S. Agriculture Department Plans to Deploy Grok Despite Safety Concerns

The U.S. Department of Agriculture intends to proceed with utilizing xAI's Grok chatbot for various applications despite ongoing concerns regarding its safety and suitability for government use. The decision includes sponsoring Grok for review through the FedRAMP program, representing a significant win for xAI in securing government contracts. Earlier reports indicated GSA coders had begun integrating Grok for Government software into broader government AI resources, though the deployment faces scrutiny over the chatbot's reliability and security standards for federal agency use.

2026-04-13

xAI CFO Departs Amid Colorado AI Law Challenge as Grok 5 Slips to Q2 2026

xAI CFO Anthony Armstrong, who was overseeing finance for both xAI and X (formerly Twitter), has left the company according to a Reuters report. Armstrong had been credited with steering X back toward financial stability following advertiser exodus. Separately, xAI filed in Colorado federal court to block a new state AI law, claiming the statute 'severely burdens' AI development and threatens Grok's operations. Grok 5, planned as a 6-trillion-parameter Mixture-of-Experts model, missed its Q1 2026 target and is now scheduled for Q2 2026.

2026-04-12

X Rolls Out Grok-Powered Auto-Translation and AI Photo Editing; xAI Reorganizes Engineering Team Ahead of SpaceX IPO

X began rolling out new Grok-powered features including automatic post translation expanding globally and an AI-powered photo editor on iOS with 'edit with words,' drawing tools, and face blurring. Separately, xAI released Grok 3 Mini via API with native function calling, JSON mode, and live search integration priced at $0.30 per million input tokens — its most affordable offering. xAI also reorganized its engineering leadership ahead of the SpaceX IPO, appointing Aman Madaan to oversee model factory and tooling and Aditya Gupta to head post-training and reinforcement learning. Security analysts noted Grok's relatively loose alignment has been cited as a potential advantage for non-standard users including the Pentagon, which reportedly began using Grok following Defense Secretary Hegseth's January 2026 announcement.

2026-04-09

Grok 4.1 Achieves 65% Hallucination Reduction with Agent Tools API; Grok 5 Targets 6T Parameters

xAI confirmed Grok 4.1 availability across grok.com, the X platform, and iOS/Android apps, with the model achieving a 65% hallucination reduction — dropping error rates from 12.09% to 4.22%. New developer capabilities include the Agent Tools API, Remote MCP Tools, Grok Voice, and Files API for building sophisticated agentic applications. Elon Musk separately stated that honest user feedback is Grok's primary development goal, with 'honest versions' maintained during training to eliminate 'bad transformers.' Grok 5, targeting 6 trillion parameters with native multimodal and video understanding, was originally scoped for Q1 2026 but has not yet shipped.

2026-04-07

Dutch Court Orders xAI's Grok to Stop Processing Sexual Content; Grok 4.20 Remains Latest Release

A Dutch court ordered X and Grok to cease processing certain sexual content, marking a notable European regulatory action against xAI's model. xAI's most recent major release remains Grok 4.20 (February 27, 2026), a high-performance reasoning model, with Elon Musk reiterating that honest user feedback is a core objective and that xAI maintains 'honest versions' of Grok during training to eliminate 'bad transformers.' Community coverage this week pivoted to adjacent AI research tools, and Grok continues to market itself on ideological differentiation via non-filtered responses on politically sensitive queries.

2026-04-06

Grok Imagine Gets Prompt-Crafting and Voice Input Enhancements; xAI Pushes into Generative Multimedia Capabilities

Elon Musk publicly showcased Grok Imagine, xAI's AI-powered image and video generation tool, with new enhancements rolled out in early April 2026, including active prompt-crafting assistance and voice input support highlighted for accessibility to young children. The tool is integrated into the X platform and tied to Tesla's AI ecosystem, signaling xAI's broader push into generative multimedia capabilities. Earlier roadmap reporting noted Grok 4.1 achieved a 65% reduction in hallucination rates (from 12.09% to 4.22%) with 1483 Elo on LMArena, and Grok 5 targets a 6-trillion-parameter architecture.

2026-04-04

More vendors

AnthropicOpenAIGoogleAWSAzureMetaNVIDIAMistralAppleHugging FaceAlibabaDeepSeekSamsung

← Browse all AI stories