🚀 Moonshot ships Kimi K3 with 2.8T parameters 🚀
China's frontier model matches Western pricing and agentic speed.

Moonshot AI has shipped Kimi K3, a 2.8-trillion-parameter multimodal model featuring a 1-million-token context window with optimizations for faster long-context decoding and agentic coding. K3 delivers competitive benchmark performance against frontier models like Claude Fable 5 and GPT-5.6 Sol, while matching Claude 5 Sonnet's pricing at $3/$15 per million tokens. In a striking demo, K3 worked autonomously for 48 hours to design and verify a chip capable of running a miniature version of itself — narrowing the competitive gap between Chinese and Western frontier providers. Open weights are scheduled for public release on July 27, giving developers worldwide free access to one of the largest models ever released.

📓 Google rebrands NotebookLM as Gemini Notebook 📓
Google deepens Gemini integration across its full product ecosystem.

Google has officially rebranded NotebookLM as Gemini Notebook, consolidating its AI-powered productivity tools under the unified Gemini brand. The rebrand comes paired with a significant functional upgrade: a new cloud computer tool that allows the platform to execute code directly, enabling more powerful data analysis workflows. Gemini Notebook now integrates across the broader Google ecosystem, including the Gemini app and Google Search, expanding distribution to millions of existing users. This move signals Google's strategic push to unify its AI product portfolio, making the notebook tool accessible through multiple touchpoints and positioning it as a central piece of the company's productivity AI stack.

⚡ NVIDIA Nemotron 3 Embed ranks #1 on RTEB ⚡
Open embedding models set new retrieval benchmarks for enterprise access.

NVIDIA has released Nemotron 3 Embed, a family of three open embedding models purpose-built for RAG, agentic retrieval, code search, and memory applications. The flagship 8-billion-parameter model has claimed the #1 overall ranking on the RTEB (Retrieval Text Embedding Benchmark), setting a new standard for embedding quality. Released as open weights, the models are freely accessible to the developer community. This strengthens NVIDIA's position in the foundational software layer powering AI retrieval infrastructure — critical for enterprise AI agent deployments. By pairing hardware dominance with top-tier open models, NVIDIA extends its competitive grip across the full AI stack.

INTERESTING TO KNOW

📉 Gemini 3.5 Pro delays sink Alphabet stock 4% 📉

Alphabet shares dropped 4% after reports that the rollout of Gemini 3.5 Pro has been delayed due to the model underperforming internal benchmarks, particularly for coding tasks. Google says the model remains in active partner testing. The report also highlights internal friction between Google's Android, DeepMind, and Cloud divisions, compounded by notable talent departures to OpenAI and Anthropic — raising questions about Google's competitive pace in the frontier model race.

🛠️ GitHub opens Copilot SDK for third-party integration 🛠️

GitHub has released the Copilot SDK, an open developer toolkit that enables engineers to embed Copilot agents directly into custom applications and third-party developer tools. This integration play positions Copilot as embeddable infrastructure rather than a standalone product, expanding its reach as a platform-level AI coding assistant. The release could accelerate adoption among enterprise teams building proprietary development environments and significantly broaden the Copilot ecosystem.