Former OpenAI CTO Mira Murati Launches Thinking Machines' Inkling: Open-Weights Multimodal AI Model with 975 Billion Parameters and Self-Training Demo
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.
Related articles
Positive Technologies Adds Unified Console to PT X for Centralized Incident and Asset Management
Positive Technologies has introduced a single console within its PT X platform that combines incident management, asset control, and security posture assessment. Customers can now monitor attack attempts around the clock, view possible causes, track response status, and prioritize events without switching between multiple interfaces. The system lets organizations predefine which assets Positive Technologies may act upon and choose between manual customer response, automatic PT X actions, or approval-based workflows, with full audit reports generated afterward. A dedicated dashboard highlights coverage gaps by showing hosts protected by MaxPatrol EDR and flagging assets without agents installed. Machine learning capabilities have been expanded through a combination of large language models and classical ML to accelerate early attack detection and automate handling of low- and medium-severity incidents. Additional features include external perimeter analysis that reveals internet-facing resources from an attacker’s perspective and a bug-bounty-style cyber testing program that rewards researchers for identifying unacceptable events.
Meta Allows Over 50 Paid Ads Sexualizing Minors to Run on Facebook, Instagram, Messenger and Threads
Meta's advertising moderation systems failed to detect and remove more than 50 paid advertisements containing sexualized images of minors across its major platforms. The ads, some generated using generative AI tools, appeared on Facebook, Instagram, Messenger, and Threads over a nine-month period. Researchers at Tech Transparency Project identified that several of these advertisements remained active into early August. The content was not organic user posts but commercially purchased ads that bypassed Meta's review processes. The incident highlights ongoing weaknesses in automated and human moderation of paid promotional material involving prohibited themes. Tech Transparency Project emphasized that the scale and persistence of the ads point to systemic gaps in Meta's enforcement mechanisms for child safety policies.
Yandex Go Develops Custom DSL to Handle Complex Dynamic Ride Pricing Logic
Yandex Go has moved its ride pricing algorithm out of the main service code into a dedicated domain-specific language to avoid frequent deployments and tangled conditional logic. The pricing system accounts for geozones, demand surges, discounts, toll roads, extra stops, and special requirements such as transporting pets or sports equipment. Parameters are collected in parallel, dependencies are modeled as a graph, and requests run asynchronously on userver so that a single slow source does not break the entire calculation. Because rules change on average twice a week, keeping the logic in C++ would require rebuilding and redeploying 50 pods for roughly 40 minutes each time. The new DSL supports conditions, functions, immutable values, and fold operations instead of loops; rules are chained so each step receives the current price and metadata and returns an updated result. ANTLR 4 defines the grammar while the Z3 verifier ensures programs cannot produce invalid prices. The platform is already used by Yandex Taxi, Delivery, and the electric-vehicle charging service.
Third Edition of The Ultimate Kali Linux Book Released with Expanded OSINT and Pentesting Coverage
The publishing house Piter has released the third international edition of The Ultimate Kali Linux Book by Glen Singh. The updated volume provides comprehensive guidance on using Nmap, Metasploit, Aircrack-ng and Empire for ethical hacking and penetration testing. Significant revisions include a new chapter on OSINT, refreshed practical exercises and clearer descriptions of virtual lab environments. The book targets both beginners and experienced IT professionals seeking to master vulnerability assessment, wireless network testing and web application exploitation. Readers learn to build testing labs, perform reconnaissance, exploit network weaknesses and evaluate corporate infrastructure security. Author Glen Singh holds an MSc and multiple certifications from EC-Council, Cisco and Check Point, bringing real-world Red Team and Blue Team experience to the material.