432 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
Mistral AI's new OCR 4 model outperformed competitors in 72% of blind test cases, according to the company. The model supports 170 languages and is available via API, Studio, and Microsoft Foundry.
Mistral AI announced Mistral OCR 4 on June 23, 2026, achieving an 85.20 score on OlmOCRBench, the top overall result among tested models.
OpenAI's GPT-5.5-Cyber model surpassed Anthropic's Mythos in cybersecurity benchmarks, scoring 85.6% on CyberGym and 39.5% on ExploitGym.
Cursor announced its first self-trained AI model, set for release in weeks, with compute usage up to 20 times higher than prior models.
ByteDance's Seedance 2.5 can generate single video clips up to 30 seconds long without post-stitching, according to the company.
IEEE offers a virtual training course on designing, securing, and deploying large language models, launching in January 2025.
Apple's Siri AI, now part of iOS 27, offers personalized recommendations and improved interaction in its beta form, as tested in San Francisco.
OpenAI introduced Record & Replay for Codex, allowing users to automate workflows by replaying tasks indefinitely. The feature is now available in macOS and requires Computer Use to be enabled.
Sakana AI's Fugu outperforms Anthropic's Fable 5 and Mythos Preview in benchmarks, despite not using those models in its pool. The system is available in two variants.
Samsung Electronics has made ChatGPT and Codex available to all employees in Korea and DX division worldwide, marking one of OpenAI’s largest enterprise deployments.
OpenAI introduced scheduled task controls in ChatGPT, allowing users to manage and automate tasks more efficiently.
Anthropic's sales rose 2.5 percentage points in May to 41% of AI subscriptions, surpassing OpenAI's 39.5% share, despite a Trump administration ban on its models.