Two halves of the same sentence.
Work backwards from the human. Thinking Machines builds the science of customizable, reliable, human-shaped intelligence. 🤫 builds the place it should live — owned, private, at the edge. Here is how the two roadmaps line up.
A proposed alignment from public information — not a claimed joint roadmap.
Their theme → what 🤫 owns.
Each row pairs a theme Thinking Machines has published with the owned, consent-first answer 🤫 has been building for five years. Their science, our place for it to live.
| Thinking Machines' theme | The 🤫 answer — owned & consent-first | Product |
|---|---|---|
| Bring intelligence to where the knowledge lives — AI must be distributed to benefit from distributed knowledge.The Future Worth Building Is Human ↗ | A personal supercomputer you own, in your home or garage, so the intelligence lives where your data already is. | 🤫 Puppy One |
| Human participation is a technical challenge — the interface must invite and reward it.Interaction Models ↗ | A private agent whose interface is consent — it shows its work, asks, and keeps a receipt for every access. | 🤫 Private Agent One |
| Decentralized alignment — values belong across many people and models, not a handful of places.The Future Worth Building Is Human ↗ | A distributed edge-supercomputing grid owned by ordinary people — decentralized ownership, not just decentralized models. | 🤫 Factory One |
| Efficient fine-tuning so anyone can teach a model their own expertise (Tinker; LoRA Without Regret).Announcing Tinker ↗ | Teach-your-own private model on compute you own — your expertise stays yours, on your hardware, by consent. | 🤫 Private Agent One + Puppy One |
| Real-time, multimodal interaction between people and machines.Interaction Models ↗ | An always-on, voice-first agent and a presence wearable for the people you love. | 🤫 Agent One + Tag One |
The frontier models — through open standards.
🤫 Agent One is provider-neutral by design. We integrate the best frontier coding and reasoning models through the open Model Context Protocol — so your agent is never locked to one vendor, and the science of teaching-your-own-model runs on compute you own.
Anthropic — Claude & Claude Code
Claude Opus 4.8, Sonnet 5, Haiku 4.5; Claude Code (CLI/IDE/web) and the Agent SDK; a first-class MCP host.
Claude models ↗OpenAI — GPT-5.6 & Codex
GPT-5.6 (Sol / Terra / Luna tiers, incl. the Pro-tier Sol Pro) and the Codex agentic coding system (CLI + IDE).
Previewing GPT-5.6 ↗xAI — Grok 4.5
grok-4.5, xAI's current flagship on the xAI API (OpenAI/Anthropic-SDK compatible).
xAI models ↗Google — Gemini
Gemini 3.1 Pro and Gemini 3.5 Flash, plus the open-source Gemini CLI.
Gemini API changelog ↗Thinking Machines — Tinker
The fine-tuning API for teaching open-weight models your own expertise — the science we most want to run on owned compute.
Announcing Tinker ↗The open standard — MCP
Model Context Protocol (spec 2025-11-25), donated to the Linux Foundation's Agentic AI Foundation and supported by Anthropic, OpenAI, Google, and Microsoft — what makes provider-neutral honest.
MCP specification ↗Model names verified as of July 2026; the durable commitment is to the frontier models via open standards, not to any single version. Naming and availability change fast — we track the primary sources.
The leadership, in the open.
The public leaders whose work maps to a 🤫 alignment — named only where they keep a public presence, each cited. Reported departures are excluded; we re-verify before any outreach.
Top leadership
Mira Murati
Former OpenAI CTO building an independent, research-first lab — the counterpart for a data-and-compute-ownership conversation.
John Schulman
OpenAI co-founder and RLHF pioneer; his post-training work maps directly to teach-your-own-model on owned edge compute.
Soumith Chintala
PyTorch co-creator and a leading open-source voice — the highest-value single alignment target for 🤫's open, own-your-compute story.
GTM & commercial
Thinking Machines has not publicly named a go-to-market or partnerships leader (they were hiring for GTM as of 2026). We will identify the right commercial counterpart through a warm, credible introduction rather than name someone unverified.
Product & engineering
Horace He
Deterministic-inference and kernel work — directly relevant to reliable inference across a distributed edge fleet.
Kevin Lu
Efficient post-training / distillation — the science of low-cost model customization on owned hardware.
Jeremy Bernstein
Training-dynamics and optimization theory — efficiency research that supports low cost per watt.
Win-win-win.
The person wins
Their expertise becomes their own private model, on their own supercomputer — leverage without surrendering their data.
Thinking Machines wins
Their science of customizable, reliable models reaches people who own the compute it runs on — distribution true to their own thesis.
The world wins
Intelligence is owned and distributed, not concentrated — decentralized alignment made durable by decentralized ownership.
Public facts, cited.
- Thinking Machines Lab — official site ↗
- Connectionism — their research blog ↗
- "The Future Worth Building Is Human" (Jul 10, 2026) ↗
- Announcing Tinker (Oct 1, 2025) ↗
- Tinker — product ↗
- Thinking Machines Lab — GitHub ↗
- Wikipedia — Thinking Machines Lab (founding & funding) ↗
- TechCrunch — $12B seed (Jul 15, 2025) ↗
This is a public study and an open, honest invitation from 🤫 hushh — an admiration of Thinking Machines' public work and a proposal to explore a partnership. It is not a claimed deal, affiliation, endorsement, or sponsorship. Every fact here is drawn from Thinking Machines' own public writing or reputable press; figures that are press-reported are labeled "reported", and quotations are verbatim from the cited source. Companies and facts change — verify against the primary sources before acting.