Sam Altman posts first-hand OpenAI product and strategy updates alongside short personal reactions and company-level statements.The account is a major primary signal for OpenAI timing and executive intent, but its brevity often creates ambiguity and outsized speculation. Use it to identify a development, then verify details elsewhere.Popular postsSol and Terra Model Launch Limited PreviewOpenAI Current Plan AnnouncementChatGPT Memory System Upgrade Rolling Out
Updated July 12, 2026
X is still one of the fastest places to see AI research, product launches, technical discussion, and reactions from the people building major models and developer tools.
We separate individuals from companies so first-hand commentary does not get buried beneath large corporate accounts. Use the individual list for judgment and technical context, and the company list for official announcements.
Best AI X Accounts to Follow
Sam Altman5.07M followersOpenAI products and strategy
Andrej Karpathy2.87M followersLLMs and AI-native software
Andrew Ng1.6M followersApplied AI education
Demis Hassabis1.1M followersDeepMind research and scientific AI
Greg Brockman988.7K followersOpenAI product and engineering updates
Ilya Sutskever706K followersAdvanced AI research and safety
Geoffrey Hinton603.1K followersAI risk and scientific impact
Boris Cherny492.4K followersClaude Code development
Jim Fan439.1K followersPhysical AI and robotics research
Dario Amodei380.5K followersAdvanced AI risk and policy
Matt Shumer367.9K followersRapid model and workflow experiments
Clement Delangue359.5K followersOpen AI ecosystem strategy
Ethan Mollick358.1K followersEvidence-based AI experimentation
Simon Willison189.4K followersPractical scrutiny of language models
swyx163.2K followersAI engineering and developer ecosystemsAndrej Karpathy publishes original technical concepts, experiments, and vocabulary for understanding how language models are changing software and research practice.This is one of the most influential individual AI feeds because short posts often introduce a useful mental model or workflow that others later adopt. The account is selective and idea-dense rather than a complete news source.Popular postsAndrej Karpathy Joins AnthropicLLM Knowledge Bases Personal Wiki SystemLLM Wiki Idea File and Agent-Driven Development
Andrew Ng uses X to share educational resources, applied AI lessons, and arguments about how companies and workers should respond to new capabilities.The account is strongest when it points to a concrete course, engineering pattern, or labor-market observation. It is lower-volume than many news feeds and regularly promotes his own education and company ecosystem.Popular postsAI Forward Deployed Engineer Role and Job MarketLoop Engineering for AI-Driven Product DevelopmentServing LLMs Efficiently Course
Demis Hassabis
DeepMind research and scientific AI1.1M followersGoogle DeepMind co-founder and CEO
Demis Hassabis posts selective first-hand updates about Google DeepMind’s models, scientific programs, people, and long-term research direction.The account is valuable for executive context around major DeepMind developments and the lab’s scientific priorities. It is not frequent or technically complete, so follow the linked research for substance.Popular postsGemma 4 150M Downloads, 12B Model ReleaseGemini 3.5 Flash Performance BenchmarksJohn Jumper Leaves DeepMind for Anthropic
Greg Brockman shares first-hand OpenAI product, model, and developer updates, often amplifying concrete use cases and technical examples.The account is useful for fast access to OpenAI engineering and product material from a co-founder. It is a promotional primary source, not an independent assessment of capability or safety.Popular postsDigitalOcean Codex Remote Session SetupGPT-5.6 Sol preview benchmark resultsCodex Use-Cases: AI Teammate for Software and Operations
Ilya Sutskever
Advanced AI research and safety706K followersAI researcher and SSI co-founder
Ilya Sutskever posts extremely selectively about advanced AI research, safety, and the direction of Safe Superintelligence.The account is consequential because of the author, not because it is a rich feed. Posts are rare and cryptic, so it should be monitored as a primary signal rather than recommended for regular learning.Popular postsAnthropic and OpenAI stance on government AI useImportant work
Geoffrey Hinton
AI risk and scientific impact603.1K followersAI researcher and professor
Geoffrey Hinton posts rarely, mainly around AI risk, scientific work, and public discussion of the consequences of advanced systems.The authority and distinct risk perspective make individual posts notable, but posts are too rare for the account to function as a regular update feed.Popular postsGeoffrey Hinton Endorses AI Risk ReportAdam Brown Lecture on AI Impact on Physics
Boris Cherny
Claude Code development492.4K followersClaude Code creator and engineering leader
Boris Cherny leads Claude Code work at Anthropic and shares concise first-hand notes about new capabilities, engineering patterns, and how the product is used internally.This is a high-signal account for Claude Code users because small posts often reveal a workflow or implementation detail before it reaches polished documentation. It is product-specific and naturally enthusiastic about Anthropic’s approach.Popular postsFable 5 Model Capabilities and Subscription DetailsSelf-Verification Loops for Extended AI Model RunsFable 5 Available in Claude Code and Cowork
Jim Fan shares research and synthesis about robotics and physical AI from his work at NVIDIA, often highlighting open systems and the path toward general-purpose embodied agents.This is one of the clearest individual feeds for tracking physical AI rather than language-model product news. Posts combine original work with thoughtful curation, though NVIDIA’s research agenda naturally shapes the selection.Popular postsRobotics Endgame: Physical AGI RoadmapENPIRE: Autonomous Robot Research in Physical WorldCaP-X Open-Source Agentic Robotics Framework
Dario Amodei posts selectively about Anthropic’s policy positions, advanced-model risks, national security, and the social consequences of rapid capability growth.The account matters because it provides direct statements from a frontier-lab CEO on policy and risk. Posts are sparse and represent Anthropic’s institutional view, so contrasting analysis is essential.Popular postsAdolescence of Technology: AI Risks to Security andAnthropic Expansion to India Claude Code GrowthPolicy on the AI Exponential
Matt Shumer
Rapid model and workflow experiments367.9K followersAI founder and product builder
Matt Shumer publishes frequent, hands-on reactions to new models and prompts from the perspective of an AI product founder.The feed is useful for fast experiments and practical ideas that can be reproduced immediately. It is highly reactive and enthusiastic, so isolated tests should not become broad model conclusions without replication.Popular postsFable Model Export Suspension ImpactClaude Fable 5 Redeployment Coding Tasks Fall Back to OpusSpeed Optimization Prompt Test Results
Clement Delangue
Open AI ecosystem strategy359.5K followersHugging Face co-founder and CEO
Clement Delangue shares a founder’s perspective on open models, decentralized AI infrastructure, and the growth and policy questions around the Hugging Face ecosystem.The feed consistently surfaces concrete open-model adoption signals and a clear counterpoint to closed frontier-lab narratives. Company advocacy is explicit, but the underlying ecosystem data is often genuinely useful.Popular postsLocal Models Answer 71.3% of Real-World QueriesConcentration of Power and Wealth as Primary AI RiskUS Government Training Open-Source AI Models
Ethan Mollick
Evidence-based AI experimentation358.1K followersWharton professor and AI researcher
Ethan Mollick publishes a very high volume of model experiments, research interpretation, and observations about how AI changes work, school, and creative production.This is among the highest-signal AI accounts on X because the posts frequently show the experiment or cite the paper behind the claim. The volume is substantial, but the feed contributes original interpretation rather than simple aggregation.Popular postsUS Government Could Ban Open Weights ModelsFable AI Poem Removing Vowels ProgressivelyClaude Code Reconstructs SimRefinery from Screenshots
Simon Willison
Practical scrutiny of language models189.4K followersDeveloper and AI researcher
Simon Willison tests language models and developer tools in public, combining implementation detail with careful attention to security, provenance, and model limitations.This is one of the highest-signal practitioner feeds because observations are usually reproducible and linked to code, logs, or source material. The account moves quickly, but the accompanying blog provides durable context.Popular postsUber caps coding agents at $1,500 monthly per employeeAnthropic Models Offline Personality ClashesFable Jailbreak Critique and Export Controls
swyx publishes a high-volume stream of AI engineering observations, developer-tool discoveries, event material, and ecosystem synthesis.The account is valuable for spotting emerging developer patterns and people before they become obvious. The volume and network-driven amplification can be noisy, so longer Latent Space work provides better context.Popular postsCursor Origin Git Competitor AnnouncementAnthropic IPO Valuation Two TrillionAI Engineer World’s Fair 2026 Reaches 6k Attendees
OpenAI4.92M followersOpenAI releases and company updates
SpaceXAI2M followersGrok models and products
Google DeepMind1.45M followersFrontier AI research
Anthropic1.33M followersClaude and Anthropic updates
Hugging Face700.6K followersOpen models and community releases
Perplexity492.2K followersAI search and agent products
Midjourney416.8K followersMidjourney product releases
LangChain251.9K followersProduction AI agents
Mistral AI183.9K followersMistral models and enterprise products
Artificial Analysis101.1K followersIndependent model and infrastructure benchmarksOpenAI’s primary X account publishes rapid first-party updates across products, research, developer tools, deployments, and corporate milestones.The account is necessary for confirming official releases and availability, particularly when products change quickly. It is a promotional source and should not determine comparative quality or policy conclusions by itself.Popular postsOpenAI Launches Deployment Company for AIDaybreak: Frontier AI for Cyber DefendersCodex Preview in ChatGPT Mobile App
xAI’s X account publishes first-party updates across Grok models, generative media, agents, APIs, and platform partnerships.This is the primary source for xAI launch timing and demonstrations, especially because releases often appear on X first. Claims are tightly tied to the company’s own platform and require independent safety and quality evaluation.Popular postsGrok Models Available on Databricks Agent BricksGrok Imagine Video 1.5 ReleaseVoice Agent Builder No-Code Platform Launch
Google DeepMind’s X account publishes a dense mix of research, model, product, and scientific-application updates from the lab.This is a strong primary source for tracking the breadth of DeepMind’s work, particularly scientific and multimodal applications. The short format often compresses research nuance, making the linked paper or blog post the real evidence.Popular postsGemma 4 12B Unified Multimodal ModelGemini Omni Multimodal Video CreationAI Agents Solve Open Mathematical Problems
Anthropic’s primary X account publishes model and product releases, research, safety positions, policy responses, and major company news.This is essential for confirming Anthropic announcements quickly and often carries more immediate context than formal documentation. It remains a company channel with a direct stake in model, safety, and policy claims.Popular postsUS Export Control Directive Suspends Fable 5 Mythos 5Claude Accelerating AI Development and Self-ImprovementClaude Opus 4.8 Release
Hugging Face
Open models and community releases700.6K followersOfficial Hugging Face account
Hugging Face’s X account is a high-volume stream of open model, dataset, tool, and community releases across many AI modalities.The breadth makes it one of the best discovery feeds for open AI, including projects too small for mainstream coverage. That same volume can be overwhelming, and the account often amplifies releases without evaluating their quality.Popular postsLFM2.5-8B-A1B Device-Optimized Model ReleaseCohere Command A+ Large Language Model25+ Open-Weight AI Model Releases Across Modalities
Perplexity’s X account publishes frequent first-party updates across its search, research, browser, and agent products.The feed is useful for keeping up with a product that adds models and workflows rapidly. The launch cadence can make incremental integrations look transformational, so test whether a feature materially improves research quality.Popular postsNemotron 3 Ultra Available for Perplexity Pro and MaxClaude Fable 5 Available in Computer as OrchestratorHarvard Research on AI Agents vs Chat Interfaces
Midjourney’s X account is a first-party release and demonstration channel for its image and video generation products.This is the best social source for timely official Midjourney version changes and visual examples. The feed is sparse and does not compare quality, rights, or workflow fit with competitors.Popular postsMidjourney V8.1 Default Model Deprecates V8
LangChain’s X account publishes a high-volume stream of agent-development patterns, product releases, examples, and conference material.This is useful for discovering current production-agent techniques and seeing what developers build in the LangChain ecosystem. The pace is noisy and vendor-specific, so follow linked code and documentation for durable detail.Popular postsLangGraph Agent to Voice Agent with PipecatAgent Development Lifecycle Production SystemsMCPs vs CLIs for Building Agents
Mistral AI
Mistral models and enterprise products183.9K followersOfficial Mistral AI account
Mistral AI’s X account publishes first-party model, product, and enterprise-use updates, including work in document processing and scientific industries.The feed provides a concise view of what Mistral is shipping and where the company is positioning its platform. Posts are relatively sparse, and technical claims require the release documentation.Popular postsMistral OCR 4 Optical Character RecognitionMistral Workflows Public Preview Enterprise AI OrchestrationMistral AI Solutions Aerospace Automotive Energy Physics
Artificial Analysis
Independent model and infrastructure benchmarks101.1K followersAI model evaluation company
Artificial Analysis publishes frequent benchmark results and market comparisons across language, image, video, speech, and infrastructure providers.This is one of the most useful data-first X feeds because posts usually contain a comparable metric rather than a vague model reaction. Rankings still depend on the benchmark design, so important conclusions require the underlying methodology.Popular postsClaude Sonnet 5 Intelligence Index Score 53AA-Briefcase Benchmark for Agentic Knowledge WorkGLM-5.2 Leading Open Weights Model Index
How to choose
Follow a few primary sources that match your work, then add independent researchers or builders who explain what a release means. High engagement is useful for discovery, but it is not evidence that a claim is correct.Other X Accounts to Consider
ElevenLabs: Generative voice and media products.
Arena.ai: Public AI model evaluation.
LlamaIndex: Data-backed AI applications.
Together AI: Open-model infrastructure and performance.
Cohere: Enterprise and sovereign AI.
Gary Marcus: Critical analysis of AI claims.
Elvis Saravia: Curated AI research for builders.
Alex Albert: Claude research and development.