Observed Signal · May 12, 2026 · Technical Release · Source: techcrunch · Impact: 2/5 · Sentiment: Neutral
Thinking Machines builds AI that listens while it talks
Thinking Machines Labs, founded by former OpenAI CTO Mira Murati, has published a technical preview describing an "interaction model" designed for full‑duplex, real‑time multimodal conversation. The model targets subsecond latency (about 0.4 seconds) so it can respond while a person is still speaking — including interrupting to correct errors, provide proactive answers, or perform simultaneous translation — and supports continuing conversation while working on background tasks. Company demos show these capabilities but the model is not yet public; Thinking Machines plans a limited research preview for testers in the coming months. Related technical disclosures (research preview) describe TML‑Interaction‑Small (a mixture‑of‑experts architecture) and benchmark wins on audio/vision timing evaluations, indicating the work is positioned as a research/technical preview rather than a commercial product.
A research milestone in conversational LLM interactivity could influence future real-time voice/chat interfaces, but it is a preview from a startup (not yet public), so immediate industry impact is limited.
Track OpenAI Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Thinking Machines Labs is developing an "interaction model" for real‑time, full‑duplex multimodal interaction.
- The model aims for ~0.4 second response times to allow interruption and simultaneous dialogue.
- Founder Mira Murati left OpenAI at the end of 2024 and leads Thinking Machines Labs.
- The startup reportedly raised more than $12 billion from investors before shipping a product.
- Thinking Machines plans a limited research‑preview release to testers in the coming months.
Connected Companies & Entities
4 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Thinking Machines Unveils Real‑Time Interaction Models
Thinking Machines Lab (Thinky), the startup founded by former OpenAI CTO Mira Murati, published a research preview of new “Interaction Models” designed for real‑time, human‑like collaboration across audio, video and text. The models use a micro‑turn architecture that processes input in ~200ms chunks and support full‑duplex simultaneous speech and listening. The company highlighted TML‑Interaction‑Small, a 276‑billion‑parameter mixture‑of‑experts model that it says responds in ~0.40 seconds and can interleave listening, tool calls (search, browsing) and UI generation. Thinking Machines raised a headline $2 billion early funding round (reported as seed/early‑stage) and lists investors and partners including Andreessen Horowitz, NVIDIA, AMD, Cisco, Jane Street and Sequoia Capital; NVIDIA also announced a 2026 partnership to deploy one gigawatt of computing capacity. The lab plans a limited research preview in coming months with a wider release later in 2026.
Mira Murati Reemerges, Previews 'Interaction Models'
Mira Murati, CEO of Thinking Machines Lab and former CTO of OpenAI, made her first major public appearance in about 18 months in a Bloomberg interview. She described Thinking Machines’ work over the past year-and-a-half — hiring, fundraising, and shipping Tinker, an API for fine-tuning open-source models — and previewed a new class of AI interfaces the company calls “interaction models.” These models are designed to process continuous streams of audio, text and video in roughly 200-millisecond intervals to better capture the flow and texture of human communication. Murati framed the work as an early step without a firm release date, addressed internal departures at her company, and reflected on governance and decision-concentration risks highlighted by OpenAI’s November 2023 leadership crisis.
Thinking Machines' Interactive Models: Model as Interface
The Sequence published an essay on May 20, 2026, examining Thinking Machines' work on interactive, multimodal models and arguing that future collaboration with AI must move beyond simple next-token text streams. The piece highlights Thinking Machines' approach to making the model itself the user interface, treating collaboration as a temporal, interactive process rather than a serialized text exchange. The essay links to a Thinking Machines video demonstration and describes the work as early-stage but impressive, positioning interactive multimodal systems as a potential evolution in how humans and models collaborate.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.