Former OpenAI CTO Mira Murati Launches Thinking Machines' Inkling: Open-Weights Multimodal AI Model with 975 Billion Parameters and Self-Training Demo

Thinking Machines, founded by former OpenAI technical director Mira Murati, has released Inkling, its first open-weights multimodal AI model that supports text, images, and audio in a unified architecture. The model uses a mixture-of-experts design with 975 billion total parameters but activates only 41 billion at once, supports a 1-million-token context window, and was trained on 45 trillion tokens spanning text, images, audio, and video. A smaller Inkling Small variant with 12 billion active parameters was also introduced for faster and cheaper inference. Key innovations include adjustable reasoning depth that lets developers control compute usage per query and a self-training experiment where the model autonomously fine-tuned itself via the Tinker platform to avoid using one letter of the English alphabet. Weights are now available on Hugging Face with support for Transformers, vLLM, SGLang, and llama.cpp, positioning Inkling as a flexible foundation for further customization rather than a direct competitor to closed frontier models.

securitylab_nJuly 16, 2026Translated from Russian

Read at source

Releasing yet another powerful language model is no longer enough to stand out. Thinking Machines, the startup founded by former OpenAI technical director Mira Murati, has chosen a different approach to attract developers. The company unveiled Inkling, its first open-weights model that can be freely used and adapted for specific tasks.

Inkling is a multimodal model capable of processing text, images, and audio without separate processing modules for each modality. It is built on a mixture-of-experts architecture containing 975 billion parameters, of which only 41 billion are active at any given time. This design significantly reduces computational costs. The model supports a maximum context length of 1 million tokens and was trained on a dataset of 45 trillion tokens that included text, images, audio, and video. Alongside the main model, the company released a preliminary Inkling Small version with 12 billion active parameters optimized for lower-cost and faster deployment.

One of Inkling’s standout features is the ability to regulate reasoning depth. Developers can choose how many computational resources the model should allocate to solving a problem. For simple queries the system responds faster and consumes fewer tokens, while for complex tasks it can increase internal computation. According to the company, in certain programming scenarios Inkling uses approximately three times fewer tokens than several other open models while maintaining comparable quality.

Thinking Machines acknowledges that Inkling does not yet aim to be the strongest model on the market. Closed models from OpenAI, Anthropic, and Google continue to lead most comprehensive benchmarks, while Chinese models remain ahead in certain disciplines. Instead of competing for top rankings, the company focused on creating a versatile foundation for subsequent training and customization for enterprise use cases.

To support further development, Thinking Machines offers its own Tinker platform. Through this platform, developers can fine-tune Inkling on their own data without building complex infrastructure themselves. To demonstrate the platform’s capabilities, the company conducted an unusual experiment: Inkling was tasked with training itself. The model autonomously created a training task, executed the process via Tinker, evaluated the results, and switched to the updated version. In the demonstration, it was given the unusual objective of learning to answer questions while completely avoiding the use of one letter of the English alphabet.

The company has also made the model weights available on Hugging Face and added support for popular inference frameworks including Transformers, vLLM, SGLang, and llama.cpp. Inkling is designed not only for cloud services but also for continued training, creation of specialized assistants, and development of autonomous agents capable of executing complex sequences of actions.

The release of Inkling represents an important development for the Western open-weights AI community. After Meta reduced its activity in this area and many organizations began turning to Chinese models, developers now have another major Western-origin solution that can be freely run, modified, and adapted to their own tasks thanks to its open weights.

OpenAI Thinking Machines Inkling Hugging Face Transformers vLLM llama.cpp

ShareX Facebook LinkedIn Reddit

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in Other

Other

Mail.ru Reduces Lifetime 1TB Cloud Storage Promo to Standard 8GB Quota for Select Users

Mail.ru has begun notifying a small group of long-term Cloud Mail users that their complimentary 1TB storage, originally granted through a 2013-2014 app installation promotion, will be reduced to the standard 8GB free tier. Affected users receive personalized deadlines, with some facing storage limits starting 30 October and final deletion of excess files scheduled for 30 January 2027. The company confirmed the change impacts fewer than 0.5 percent of Cloud Mail users and stated that core free features remain available. Once the 8GB limit is exceeded, uploads, edits, and sharing are blocked while users have three months to download data or purchase additional space before files move to the trash for 14 days prior to permanent removal. This follows a similar reduction of lifetime 100GB storage performed by the service in 2024.

AntiMalwareOct 29, 2026Translated from Russian

Other

Kimi K3 Open Model Surpasses Closed AI Leaders with Million-Token Context and 896 Experts

Kimi K3 is an open-weights model that outperforms closed systems from GPT and Anthropic on several independent benchmarks. Developers achieved this by scaling three dimensions simultaneously: one million tokens of context, deeper layers, and 896 Mixture-of-Experts specialists. The team replaced the standard KV-cache with a fixed-size state matrix updated via the delta-rule, solving numerical instability by clamping the forgetting coefficient alpha between e^-5 and 1. Sequential state updates were parallelized across chips, GPUs, and multi-GPU nodes using the open-source FlashKDA kernel built on CUTLASS. Inter-layer information flow was preserved through learned pseudo-queries that selectively retrieve relevant outputs from prior blocks rather than passing full activations. Load balancing across 896 routed experts uses per-expert quantile thresholds and a distributed histogram method to guarantee fair token allocation without full sorting.

HabrOct 29, 2026Translated from Russian

Other

Innostage Completes Functional Testing of PT NGFW 1.10.3 Across 156 Criteria

Innostage has published the second part of its evaluation of PT NGFW version 1.10.3, focusing on functional capabilities rather than raw performance. The testing covered 156 individual checks grouped into categories such as basic technical features, firewalling, IDS/IPS, web filtering, malware protection, logging, routing, deployment, cryptographic controls and access management. Out of the total, 136 checks passed fully, seven passed with remarks, and 13 did not meet the predefined criteria. Key confirmed functions include high-availability clustering, centralized management via PT-MNGT, AD-based rule application, Site-to-Site VPN, TLS inspection, and Syslog integration with external security systems. Several advanced scenarios, including custom block pages, application risk classification, and centralized software distribution, were not confirmed under the test conditions. The report provides detailed protocol notes that help organizations map product capabilities against their specific infrastructure and regulatory requirements.

HabrOct 29, 2026Translated from Russian