Google Introduces Gemini 4 Argon for Long-Running Tasks

Morning, {{first name | folks}}! AI had another busy stretch. Google unveiled a new model built for long-running work, OpenAI moved against a large-scale model distillation effort, GitHub brought multi-model routing into Copilot, ElevenLabs hit a $22B valuation, and Cohere launched new models for enterprise search.

Google has introduced Gemini 4 Argon, a new frontier model designed for complex coding, enterprise work, financial research, legal drafting, and cybersecurity. It supports a 1-million-token context window, allowing it to work through large amounts of information across longer, multi-step tasks.

Argon can also find, validate, and patch critical software vulnerabilities autonomously. Google is initially making it available to trusted cyber defenders through its Fairwind Program while it continues testing the model before a wider release.

OpenAI says it disrupted a coordinated campaign that attempted to extract protected reasoning from its models at scale. The activity began on July 1 and included more than 16,000 requests across over 4,000 users during spikes on July 24 and 25, with related activity later traced across a cluster of more than 15,000 users.

OpenAI says the operators manipulated model interactions to reproduce hidden reasoning without breaking encryption, accessing databases, or directly obtaining stored conversations. The company attributes a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi, and says it has since strengthened technical controls, restricted accounts and shared its findings with industry and government partners.

GitHub is expanding HydraFusion beyond its CLI preview, bringing the multi-model system to VS Code and the Copilot app. Instead of sending every task to the same model, HydraFusion can choose between models depending on the work involved.

It can switch to a stronger model for harder tasks, use one model to review another’s work, or combine several models in the same workflow. GitHub is rolling it out as a research preview to Copilot Pro, Pro+, Business, and Enterprise users.

ElevenLabs has reached a $22 billion valuation, nearly doubling its value in seven months as demand for AI voice agents continues to grow. The company reached the new valuation through a $300 million tender offer that lets employees and existing shareholders sell their shares.

ElevenLabs says its voice agents now handle more than 15 million conversations a week, up threefold since February, across tasks including refunds, insurance renewals, and appointment bookings. The company’s technology supports voice interactions and translation in more than 90 languages.

Cohere has launched Embed 5, a new family of embedding models designed for enterprise search, RAG, and agent workflows. The models come in Pro and Fast tiers, support text, images, and mixed text-image inputs, and cover more than 100 languages with a 128K-token context window.

Embed 5 Pro scored 85.8 on Cohere’s ViDoRe V3 retrieval benchmark, ahead of Voyage 4 Large and Gemini Embedding 2. The Fast version is built for lower-latency, high-volume workloads and costs $0.08 per million tokens, compared with $0.12 for Pro. Both models use the same embedding space, so companies can index with Pro and switch to Fast for live queries without rebuilding their indexes.