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
How a Personal Site Counts Bots Without Storing Any IPs: Pure PHP, No Database, Strict CSP and 160 Tests
A personal website built entirely in pure PHP without frameworks or databases demonstrates advanced defensive security practices while tracking bot activity from an SSH honeypot. The site stores all content in a single JSON file with atomic writes and automatic backups, eliminating database-related attack surfaces and simplifying deployment to folder copying. Strict Content-Security-Policy rules prohibit inline styles and scripts without nonces, forcing the use of SVG attributes, CSSOM updates, and JSON data blocks instead of direct attribute injection. Over 160 automated tests written in PHP attack the site with XSS payloads, CSRF attempts, path traversals, and malformed Markdown to verify that no malicious output reaches the browser. Bot identification relies on behavioral signatures derived from the first twelve cleaned commands of each session, combined with HASSH fingerprints and activity patterns, allowing the site to report unique bot families and address counts without ever publishing IPs. Additional browser-based tests using Chromium validate accessibility, performance budgets, and CSP compliance across desktop and mobile viewports. The entire system runs with zero external dependencies and integrates tests directly into the deployment pipeline.
Russian Finance Ministry Seeks Central Bank Reports on Systematic P2P Transfers to Tax Authorities
The Russian Ministry of Finance has proposed amendments requiring the Central Bank to share data with the Federal Tax Service on individuals whose bank accounts show signs of regular income received from other private citizens. The changes target article 85 of the Tax Code and form part of the budget package bill numbered 617-9. Under the new rules, the Central Bank would identify such citizens through transaction analysis and transmit the information according to a future agreement with the tax service. The Federal Tax Service would also gain expanded powers to request bank statements from non-entrepreneurs without needing prior debt collection or audit triggers. If approved, the provisions would take effect on 1 January 2027 and apply to operations conducted after 1 January 2026. Separate bills under discussion would impose monthly transfer limits of 30,000 rubles on suspected clients starting in 2028 and require the tax service to notify the Central Bank about account openings and income categories.
Baidu Launches Ciyuan Exchange to Establish Trust Standards for AI Model Trading
Baidu Intelligent Cloud has introduced the Ciyuan Exchange, a dedicated trading platform for large AI models that aims to solve quality verification and credit issues in an increasingly crowded market. The platform, unveiled at the Baidu World 2026 Conference, already hosts over 100 supply resources with daily calls exceeding 50 million and daily token usage surpassing 700 billion. Founder Li Yanhong positioned the exchange as an infrastructure layer that moves beyond simple model aggregation toward verifiable identity, ongoing performance monitoring, and supplier credit records. The initiative addresses real procurement risks where identical models from different providers show up to twofold differences in throughput and reliability, according to a China Academy of Information and Communications Technology report. By combining model fingerprinting, capability testing, periodic audits, and continuous credit scoring, the platform seeks to create transparent market mechanisms similar to commodity exchanges. This development signals a broader industry shift from raw model capability competition toward standardized service quality, pricing transparency, and reliable delivery infrastructure.
Windows and MacBook Pro Setup Combines Local Network Access with Dual VPNs and Kill Switch Protection
A user detailed how they integrated an older MacBook Pro into an existing Windows workstation environment while preserving separate user accounts, file sharing, and independent VPN connections. The setup required careful routing configuration to keep the Kill Switch active on both Amnezia VLESS and AmneziaWG without breaking local SMB connectivity. After initial failures caused by the Windows Kill Switch blocking the 192.168.1.0/24 subnet, split tunneling was implemented to route local traffic directly. Separate macOS accounts were created for daily use and a dedicated “Serverok” account for remote scripting and compute tasks. SSH and SFTP ultimately replaced problematic SMB shares for the secondary account, enabling reliable file transfers at over 50 MB/s. The configuration allows simultaneous or sequential use of both machines while maintaining strong VPN protection on each.