1<!DOCTYPE html><html lang="en" class="no-js"><head><meta charset="UTF-8"><meta name="viewport" content="width=device-width, initial-scale=1.0">
1<script> 2 document.documentElement.classList.remove('no-js'); 3 document.documentElement.classList.add('js'); 4 </script>
4<meta name="author" content="Thinking Machines Lab"><meta name="description" content="Connectionism: Research Blog by Thinking Machines Lab"><link rel="apple-touch-icon" sizes="180x180" href="/images/apple-touch-icon.png"><link rel="icon" type="image/png" sizes="32x32" href="/images/favicon-32x32.png"><link rel="icon" type="image/png" sizes="16x16" href="/images/favicon-16x16.png"><link rel="preload" href="/fonts/gt-america/GT-America-Standard-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous"><link rel="preload" href="/fonts/gt-america/GT-America-Standard-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous"><link rel="preload" href="/fonts/gt-america/GT-America-Standard-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous"><link rel="preload" href="/fonts/chakra-petch/chakra-petch-v12-latin-regular.woff2" as="font" type="font/woff2" crossorigin="anonymous"><link rel="preload" href="/fonts/chakra-petch/chakra-petch-v12-latin-500.woff2" as="font" type="font/woff2" crossorigin="anonymous"><link rel="preload" href="/fonts/chakra-petch/chakra-petch-v12-latin-600.woff2" as="font" type="font/woff2" crossorigin="anonymous"><meta name="keywords" content="thinky thinkingmachines machine learning deep learning ai"><link rel="stylesheet" href="/katex/katex.min.css?v=f2f08c0551dd"><link rel="canonical" href="https://thinkingmachines.ai/"><link rel="alternate" type="application/rss+xml" href="https://thinkingmachines.ai/index.xml" title="Thinking Machines Lab"><meta itemprop="name" content="Thinking Machines Lab"><meta itemprop="description" content="Connectionism: Research Blog by Thinking Machines Lab"><meta itemprop="dateModified" content="2026-08-31T11:19:20-07:00"><link rel="stylesheet" href="/css/common.css?v=c32b5950996e" fetchpriority="high"><link rel="stylesheet" href="/css/ui.css?v=7c58697c7635"><link rel="preload" as="style" href="/css/typography.css?v=fbb994bc49ca"><link rel="stylesheet" href="/css/typography.css?v=fbb994bc49ca" media="print" onload="this.media='all'"><noscript><link rel="stylesheet" href="/css/typography.css?v=fbb994bc49ca"></noscript><link rel="preload" as="style" href="/css/fonts.css?v=446f09d40630"><link rel="stylesheet" href="/css/fonts.css?v=446f09d40630" media="print" onload="this.media='all'"><noscript><link rel="stylesheet" href="/css/fonts.css?v=446f09d40630"></noscript><link rel="preload" as="style" href="/css/syntax.css?v=3da135297e59"><link rel="stylesheet" href="/css/syntax.css?v=3da135297e59" media="print" onload="this.media='all'"><noscript><link rel="stylesheet" href="/css/syntax.css?v=3da135297e59"></noscript><title>Thinking Machines Lab</title><meta property="og:title" content="Thinking Machines Lab"><meta property="og:description" content="Connectionism: Research Blog by Thinking Machines Lab"><meta property="og:type" content="website"><meta property="og:url" content="https://thinkingmachines.ai/"><meta property="og:site_name" content="Thinking Machines Lab"><meta property="og:image" content="https://thinkingmachines.ai/images/home.png"><meta property="og:image:alt" content><meta name="twitter:card" content="summary_large_image"><meta name="twitter:title" content="Thinking Machines Lab"><meta name="twitter:description" content="Connectionism: Research Blog by Thinking Machines Lab"><meta name="twitter:image" content="https://thinkingmachines.ai/images/home.png"><meta name="twitter:image:alt" content><style>.post-group[data-astro-cid-jcsiju5y]{max-width:660px;padding-top:1.6rem;list-style-type:none;& li{padding-bottom:.4rem;position:relative}& li+li[data-astro-cid-jcsiju5y]{padding-top:.4rem}& li+li[data-astro-cid-jcsiju5y]:before{content:"";border-top:1px solid color-mix(in srgb, var(--fg4) 10%, transparent);pointer-events:none;position:absolute;top:0;left:0;right:0}& .post-item-link{color:inherit;cursor:pointer;border-radius:8px;margin:0 -1.6rem;padding:1rem 1.6rem;line-height:1.4;text-decoration:none;display:block;@media (width>=768px){grid-template-columns:7rem 1fr;align-items:baseline;gap:1.5rem;margin:0 -1.5rem;padding:1.2rem 1.5rem;display:grid}}& .post-info{flex-direction:column;align-items:flex-start;gap:.25rem;display:flex;@media (width>=768px){grid-area:1/2}}& .post-title{color:var(--fg);text-wrap:balance;text-underline-offset:2px;font-size:17px;font-weight:500;text-decoration:underline #0000;transition:color .2s,text-decoration-color .2s;
4display:inline-block}& .post-subtitle{color:var(--fg1);font-size:.95em;line-height:1.4;transition:color .2s}& .author-date{color:var(--fg4);font-size:15px;font-family:var(--font-sans);display:block}& time{font-size:15px;font-family:var(--font-sans);color:var(--fg4);margin-top:.25rem;transition:color .2s;display:block;@media (width>=768px){text-align:left;white-space:nowrap;grid-area:1/1;margin-top:0}}@media (hover:hover){& .post-item-link:hover{& .post-title{color:var(--fg);text-decoration-color:color-mix(in srgb, var(--fg2) 20%, transparent)}& .post-subtitle{color:var(--fg)}& .author-date,& time{color:var(--fg3)}}}} 5svg[data-astro-cid-t4uphou2]{cursor:none}@media (width<=640px){svg[data-astro-cid-t4uphou2]{aspect-ratio:4/3}} 6.btn[data-astro-cid-ekguhzzh]{appearance:none;min-height:40px;color:var(--fg);font-family:var(--font-sans);cursor:pointer;transition:transform .45s cubic-bezier(.34,1.56,.64,1);transition-timing-function:linear(0, .44 10%, .94 20%, 1.32 30%, 1.5 40%, 1.28 50%, .94 60%, .75 70%, .82 78%, .96 86%, 1.03 93%, 1);border:none;border-radius:100px;justify-content:center;align-items:center;gap:.5rem;padding:0 18px;font-size:15px;font-weight:400;line-height:1;text-decoration:none;display:inline-flex;position:relative;overflow:hidden}.btn[data-astro-cid-ekguhzzh]:before{content:"";opacity:0;pointer-events:none;border-radius:inherit;background:#000;transition:opacity .18s;position:absolute;inset:0}.btn[data-astro-cid-ekguhzzh]:active{transition:transform .12s cubic-bezier(.4,0,.2,1);transform:scale(.96)}.btn[data-astro-cid-ekguhzzh][disabled]{opacity:.6;cursor:not-allowed}.btn[data-astro-cid-ekguhzzh].primary{color:#000;background:linear-gradient(#f3f2f2 0%,#e6e6e6 87.5%)}.btn[data-astro-cid-ekguhzzh].primary:hover:before{opacity:.04}.btn[data-astro-cid-ekguhzzh].primary:active:before{opacity:.08}.btn[data-astro-cid-ekguhzzh].plain{color:var(--fg);background:0 0}.btn[data-astro-cid-ekguhzzh].plain:before{display:none}.btn[data-astro-cid-ekguhzzh].plain:hover,.btn[data-astro-cid-ekguhzzh].plain:focus-visible{background:#0000000d}.btn[data-astro-cid-ekguhzzh].plain:active{background:#00000014}.btn-with-chevron[data-astro-cid-ekguhzzh]{gap:.25rem}.btn-with-chevron[data-astro-cid-ekguhzzh] svg[data-astro-cid-ekguhzzh]{width:16px;height:16px;transition:transform .2s ease-in-out}.btn-with-chevron[data-astro-cid-ekguhzzh] svg[data-astro-cid-ekguhzzh] path[data-astro-cid-ekguhzzh]{opacity:.4;transition:opacity .2s ease-in-out}.btn-with-chevron[data-astro-cid-ekguhzzh]:hover svg[data-astro-cid-ekguhzzh]{transform:translate(3px)}.btn-with-chevron[data-astro-cid-ekguhzzh]:hover svg[data-astro-cid-ekguhzzh] path[data-astro-cid-ekguhzzh]{opacity:.7}@media (prefers-reduced-motion:reduce){.btn[data-astro-cid-ekguhzzh],.btn[data-astro-cid-ekguhzzh]:active,.btn-with-chevron[data-astro-cid-ekguhzzh] svg[data-astro-cid-ekguhzzh]{transition:none}} 7</style></head><body><div id="wrapper"><header id="header"><!-- Navbar for all pages --><div id="site-header" class="site-header"><div class="navbar-spacer"></div><div class="navbar-logo navbar-logo-home" aria-hidden="true"><a href="/">THINKING MACHINES</a></div><!-- Navigation group on the right --><div class="nav-group"><nav class="top-nav"><a href="/inkling/">Inkling</a><a href="/tinker/">Tinker</a><a href="/blog/">Connectionism</a><a href="/news/">News</a><a href="/#join-us">Join us</a></nav><button id="menu-toggle" class="hamburger" aria-label="Open menu" aria-expanded="false" aria-controls="mobile-menu" type="button"><span></span><span></span><span></span></button></div></div><!-- Mobile overlay menu --><div id="mobile-menu" class="mobile-menu-overlay" aria-hidden="true" hidden><nav class="mobile-menu-nav" role="navigation" aria-label="Mobile"><a href="/">Home</a><a href="/inkling/">Inkling</a><a href="/tinker/">Tinker</a><a href="/blog/">Connectionism</a><a href="/news/">News</a><a href="/#join-us">Join us</a></nav></div><div role="banner" class="logo-container"><a class="logo" href="/">Thinking<br>Machines</a><a class="home-announcement" href="/inkling/" aria-label="Inkling: Our Open-Weights Models"><span class="pill">NEW</span><span class="announcement-text">Inkling: Our Open-Weights Models</span></a></div></header><main id="main" class="index content"><article class="content"><p>Thinking Machines Lab is an artificial intelligence research and product company. Weâre building a future where everyone has access to the knowledge and tools to make AI work for their unique needs and goals.</p> 8<p>While AI capabilities have advanced dramatically, key gaps remain. The scientific communityâs understanding of frontier AI systems lags behind rapidly advancing capabilities. Knowledge of how these systems are trained is concentrated within the top research labs, limiting both the public discourse on AI and peopleâs abilities to use AI effectively. And, despite their potential, these systems remain difficult for people to customize to their specific needs and values. To bridge the gaps, weâre building Thinking Machines Lab to make AI systems more widely understood, customizable and generally capable.</p> 9<p>We are scientists, engineers, and builders whoâve created some of the most widely used AI products, including <a href="https://chatgpt.com">ChatGPT</a> and <a href="https://character.ai/">Character.ai</a>, open-weights models like <a href="https://mistral.ai/">Mistral</a>, as well as popular open source projects like <a href="https://pytorch.org/">PyTorch</a>, <a href="https://github.com/openai/gym">OpenAI Gym</a>, <a href="https://github.com/facebookresearch/fairseq">Fairseq</a>, and <a href="https://github.com/facebookresearch/segment-anything">Segment Anything</a>.</p> 10<h2 style="text-align: center;"> Science is better when shared</h2> 11<p><em>Scientific progress is a collective effort.</em> We believe that weâll most effectively advance humanityâs understanding of AI by collaborating with the wider community of researchers and builders. We plan to frequently publish technical blog posts, papers, and code. We think sharing our work will not only benefit the public, but also improve our own research culture.</p> 12<h2 style="text-align: center;"> AI that works for everyone</h2> 13<p><em>Emphasis on human-AI collaboration.</em> Instead of focusing solely on making fully autonomous AI systems, we are excited to build multimodal systems that work with people collaboratively.</p> 14<p><em>More flexible, adaptable, and personalized AI systems.</em> We see enormous potential for AI to help in every field of work. While current systems excel at programming and mathematics, weâre building AI that can adapt to the full spectrum of human expertise and enable a broader spectrum of applications.</p> 15<h2 style="text-align: center;"> Solid foundations matter</h2> 16<p><em>Model intelligence as the cornerstone.</em> In addition to our emphasis on human-AI collaboration and customization, model intelligence is crucial and we are building models at the frontier of capabilities in domains like science and programming. Ultimately, the most advanced models will unlock the most transformative applications and benefits, such as enabling novel scientific discoveries and engineering breakthroughs.</p> 17<p><em>
17Infrastructure quality as a top priority.</em> Research productivity is paramount and heavily depends on the reliability, efficiency, and ease of use of infrastructure. We aim to build things correctly for the long haul, to maximize both productivity and security, rather than taking shortcuts.</p> 18<p><em>Advanced multimodal capabilities.</em> We see multimodality as critical to enabling more natural and efficient communication, preserving more information, better capturing intent, and supporting deeper integration into real-world environments.</p> 19<h2 style="text-align: center;"> Learning by doing</h2> 20<p><em>Research and product co-design.</em> Products enable iterative learning through deployment, while great products and research strengthen each other. Products keep us grounded in reality and guide us to solve the most impactful problems.</p> 21<p><em>Empirical and iterative approach to AI safety.</em> The most effective safety measures come from a combination of proactive research and careful real-world testing. We plan to contribute to AI safety by (1) maintaining a high safety barâpreventing misuse of our released models while maximizing usersâ freedom, (2) sharing best practices and recipes for how to build safe AI systems with the industry, and (3) accelerating external research on alignment by sharing code, datasets, and model specs. We believe that methods developed for present day systems, such as effective red-teaming and post-deployment monitoring, provide valuable insights that will extend to future, more capable systems.</p> 22<p><em>Measure what truly matters.</em> Weâll focus on understanding how our systems create genuine value in the real world. The most important breakthroughs often come from rethinking our objectives, not just optimizing existing metrics.</p> 23<h2 id="join-us" style="text-align: center;"> Join us</h2> 24<p>Weâre building AI systems that push technical boundaries while delivering real value to as many people as possible. Our team combines rigorous engineering with creative exploration, and weâre looking for collaborators to help shape this vision.</p> 25<p>Follow us on X at <a href="https://x.com/thinkymachines">@thinkymachines</a> for updates or see our open roles.</p> 26<div style="margin: 2rem 0;"><a class="btn primary" href="https://jobs.ashbyhq.com/ThinkingMachines" target="_blank" rel="noopener" data-astro-cid-ekguhzzh="true">See open roles</a></div></article></main><footer id="footer"><div class="footer-columns"><div class="footer-links"><div class="footer-column"><h4>Models</h4><ul><li><a href="/news/introducing-inkling/">Inkling</a></li><li><a href="/news/inkling-small/">Inkling-Small</a></li><li><a href="/blog/interaction-models/">Interaction</a></li></ul></div><div class="footer-column"><h4>Tinker</h4><ul><li><a href="/tinker/">Overview</a></li><li><a href="https://tinker-docs.thinkingmachines.ai/tinker/">Docs</a></li></ul></div><div class="footer-column"><h4>Company</h4><ul><li><a href="/blog/the-future-worth-building-is-human/">About</a></li><li><a href="/blog/">Connectionism</a></li><li><a href="/news/">News</a></li><li><a href="https://jobs.ashbyhq.com/ThinkingMachines">Careers</a></li></ul></div><div class="footer-column"><h4>Legal</h4><ul><li><a href="/legal/terms/">Terms of service</a></li><li><a href="/legal/privacy/">Privacy notice</a></li></ul><h4 class="footer-subheading">Security</h4><ul><li><a href="/security/disclosure-policy/">Disclosure policy</a></li></ul></div><div class="footer-column"><h4>Connect</h4><ul><li><a href="https://x.com/thinkymachines">X</a></li><li><a href="https://www.linkedin.com/company/thinkingmachinesai/">LinkedIn</a></li><li><a href="https://www.youtube.com/@thinkymachines">YouTube</a></li></ul></div></div></div><div class="footer-copyright"><span>Thinking Machines Lab © 2026</span></div></footer></div>
26<script src="/js/script.min.79c2d79958493c8b72e12b13fa5d89ef49e3ae95a7aea5b82dbcc7e382c01b5a.js" integrity="sha256-ecLXmVhJPIty4SsT+l2J70njrpWnrqW4LbzH44LAG1o=" defer></script>
26<script type="module" src="https://static.cloudflareinsights.com/beacon.min.js/v31edd6df95cf4e85bb4c19e7a9bdbcba1788362987495" integrity="sha512-iIg7k2xntmwu6/uSb5tpc/hySgZc4eoL31yB29W6tJFo2akwjPWcEqnCEdJvGexCL0KEQwVYv5BlowfhVz26hg==" data-cf-beacon='{"version":"2024.11.0","token":"cda777236064450ebb47f38aeb8df789","spa":2}' crossorigin="anonymous"></script>
26 27</body></html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.