Thinking Machines Lab
PulseAugur coverage of Thinking Machines Lab — every cluster mentioning Thinking Machines Lab across labs, papers, and developer communities, ranked by signal.
- founded by Mira Murati 100%
- founded Inkling 95%
- instance of Inkling 95%
- developed Inkling-Small 95%
- instance of Inkling 2 Small 95%
- developed Inkling 2 Small 95%
- developed by Inkling-Small 95%
- developed Inkling Small 95%
- developed by Inkling 90%
- founded Mira Murati 90%
- founded Lilian Weng 90%
- developed Thinking machines 90%
- 2026-09-17 hiring Yao Xingcheng, former head of pre-training for Tencent's Hunyuan model, has joined Thinking Machines Lab. source
- 2026-06-05 product_launch Mira Murati previewed "interaction models" from Thinking Machines Lab, designed for continuous real-time processing of audio, text, and video. source
- 2026-05-17 product_launch Thinking Machines Lab launched its first AI model. source
- 2026-05-17 product_launch Thinking Machines Lab launched its first AI model focused on human-AI collaboration. source
- 2026-05-15 product_launch Thinking Machines Lab previewed new "interaction models" designed for continuous human communication.
- 2026-05-12 research_milestone Thinking Machines Lab unveiled its vision for full-duplex conversational AI, which closely resembles prior work from 面壁智能. source
- 2026-05-12 product_launch Thinking Machines Lab introduced interaction models, a new architecture for real-time human-AI collaboration. source
- 2026-05-12 product_launch Thinking Machines Lab introduced interaction models, a new type of AI for real-time collaboration. source
- 2026-05-12 product_launch Thinking Machines Lab announced its new interaction models, featuring full-duplex AI capabilities.
- 2026-05-12 product_launch Thinking Machines Lab announced its first AI model, focusing on real-time interaction. source
- 2026-05-12 product_launch Thinking Machines Lab announced a research preview of its new "interaction models" designed for real-time, simultaneous AI-human collaboration. source
- 2026-05-12 product_launch Thinking Machines Lab announced its new interaction models enabling simultaneous input processing and response generation.
8 day(s) with sentiment data
Thinking Machines Lab releases Inkling 2 Small weights under Apache 2.0 license within 7 days
Multiple sources indicate that Thinking Machines Lab intends to release the weights for Inkling 2 Small under the Apache 2.0 license next week, pending safety testing. This suggests a high likelihood of public availability of the model weights in the near future.
Inkling 2 Small shows strong performance in coding and enterprise benchmarks
Inkling 2 Small has demonstrated notable performance across several benchmarks, including SWE-Bench Pro v2 (86.3% on Hard) and EnterpriseBench-CoreCraft (71.8%, ranking third). This indicates a strong capability for agentic tasks, particularly in code generation and enterprise applications.
Inkling 2 Small multimodal capabilities include 1M context window and image input
The release of Inkling 2 Small highlights its multimodal nature, supporting both text and image inputs. Furthermore, it offers an exceptionally large context window of 1 million tokens, positioning it as a powerful tool for complex tasks requiring extensive information processing.
Thinking Machines Lab will release a specialized Inkling fine-tune for enterprise agentic workflows within 90 days.
The integration of Inkling into Databricks' Unity AI Gateway, specifically mentioning optimization for 'coding, agentic reasoning, and multimodal inputs' and connectivity with coding agents, indicates a strong enterprise focus. This suggests Thinking Machines Lab will likely release a version of Inkling specifically fine-tuned for these enterprise agentic workflows to capitalize on this early integration.
Inkling's 1M token context window is a key differentiator for long-context reasoning.
The Inkling model boasts an exceptionally large 1 million token context window, trained on 45 trillion tokens. This is a significant feature that enables advanced reasoning over extended inputs across text, image, and audio. This capability is likely to be a major draw for applications requiring deep understanding of lengthy documents or complex multimodal sequences.
-
AI market volatility: LLM prices fluctuate, GPU rentals rise, and funding news emerges · 1 source tracked
TANDIU's weekly data notes for November 2-9, 2026, highlight significant shifts in AI pricing and infrastructure. The LLM API list price for a specific model doubled and then halved within a week, underscoring the volat…
-
Thinking Machines Lab releases Inkling 2 Small with 1M context window
Thinking Machines Lab has released Inkling 2 Small, a 594B-parameter mixture-of-experts model featuring image input and a 1M-token context window. While priced competitively per token, its extensive output token usage f…
-
Inkling 2 Small: A Developer's Checklist Before Committing to a New Model
Thinking Machines Lab has released Inkling 2 Small, a 594B parameter model with 17B active parameters per token, scoring 31 on the Artificial Analysis Intelligence Index. Currently, access is limited to the Tinker platf…
-
Thinking Machines Lab launches Inkling 2 Small for enterprise agents
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released Inkling 2 Small, a 594 billion parameter model designed for enterprise agents. This model specializes in tasks such as code generation and fi…
-
Inkling 2 Small multimodal MoE model released on Tinker, Apache weights expected
Thinking Machines Lab has released Inkling 2 Small, a multimodal Mixture-of-Experts model, available on the Tinker platform. The model boasts a 594B total parameter size with 17B active parameters and supports a 256K co…
-
Thinking Machines Lab releases Inkling 2 Small with 1M context and image input
Thinking Machines Lab has launched Inkling 2 Small, a new mixture-of-experts model. This model boasts 594 billion total parameters with 17 billion active parameters, and supports a 1 million token context window along w…
-
Thinking Machines Lab releases Inkling 2 Small MoE model
Thinking Machines Lab has launched Inkling 2 Small, a 594 billion parameter Mixture-of-Experts model with 17 billion active parameters. This multimodal model accepts text and image inputs and supports a 1 million token …
-
Thinking Machines releases Inkling 2 Small with strong agentic capabilities · 2 sources tracked
Thinking Machines Lab has released Inkling 2 Small, a new Mixture-of-Experts model with 594 billion total parameters and 17 billion active parameters. The model shows significant improvements over its predecessor, parti…
-
OpenRouter offers free LLM access with request limits, not token caps
OpenRouter is offering free access to a selection of large language models in 2026, but with request limits rather than token caps. Developers can access these models with no charge per token, though daily request limit…
-
Cognition launches cheaper, faster SWE-2 Flash coding model
Cognition has released SWE-2 Flash, a more efficient version of its Devin coding model, designed for faster and cheaper execution of common coding tasks. This new model, based on the Inkling-Small architecture, utilizes…
-
LLM evaluation flawed by minor weight changes and truncation, study finds
A recent analysis revealed that even minor variations in model weights, as small as 0.005%, can lead to significant score differences of up to 1.4%p in LLM evaluations. The study found that repeating the same evaluation…
-
LLM temperature 0 is not deterministic; research explains why
LLM developers and users face challenges with achieving deterministic outputs, even when setting the temperature parameter to 0. Research from Thinking Machines Lab indicates that variations in LLM responses, even at te…
-
Paper argues deterministic LLM inference kills cognition and safety
A new paper argues against deterministic inference in large language models, asserting that it hinders the ability to model uncertainty, suppress emergent abilities, and weaken safety alignment. The authors propose 'Sto…
-
Former Tencent AI Lead Yao Xingcheng Joins Thinking Machines Lab
Yao Xingcheng, the former head of pre-training for Tencent's Hunyuan large language model, has joined the North American startup Thinking Machines Lab. This move comes after Yao Xingcheng left Tencent in July and was re…
-
LLM Fine-Tuning Methods: SFT, LoRA, QLoRA, RFT, and Distillation Explained
The article outlines various methods for fine-tuning large language models, focusing on practical applications and tool choices. Supervised Fine-Tuning (SFT) is presented as a starting point, requiring labeled input-out…
-
Chinese open-source AI models gain traction in US businesses · 1 source tracked
Chinese open-weight AI models are gaining traction with U.S. businesses, challenging the dominance of American companies like OpenAI and Anthropic. These models, such as Moonshot's Kimi K3 and Z.AI's GLM-5.3-Flash, offe…
-
Chinese labs lead open-source AI model frontier, surpassing US releases
The open-source AI model landscape is rapidly evolving, with a notable shift in the frontier of model development. Chinese labs are increasingly releasing larger and more performant models, often surpassing those from A…
-
Thinking Machines Lab releases Inkling-Small open-source MoE model
Thinking Machines Lab has released Inkling-Small, an open-source Mixture of Experts (MoE) model. Despite its smaller size, Inkling-Small surpasses its predecessor in coding tests and offers native multimodality with a 1…
-
Thinking Machines releases Inkling-Small, outperforming larger predecessor
Thinking Machines Lab has launched Inkling-Small, a new open-weights multimodal model that prioritizes efficiency over sheer size. Despite being significantly smaller than its predecessor, Inkling, Inkling-Small demonst…
-
Lilian Weng returns to OpenAI to lead self-evolving AI team
Lilian Weng, a former VP of Research and Safety at OpenAI, has returned to the company to lead its "self-evolving" (RSI) team, a high-priority research initiative. Weng had previously left OpenAI to co-found Thinking Ma…