Skip to content
Source · Daily Brief

AI Daily Brief — 16 July 2026

China’s Moonshot AI launched Kimi K3 today, a 2.8-trillion-parameter open-weight model that benchmarks near the closed frontier and claimed the top Frontend Code Arena ranking ahead of Claude Fable 5; open weights are promised by 27 July. In the same 24 hours, xAI managed simultaneous crises over Grok-generated CSAM and a credential-exfiltrating tool, while the EU issued binding orders for Google to open Android to rival AI assistants.

Top stories

  • Kimi K3 launches as China’s largest open-weight model to date. Moonshot AI’s new 2.8-trillion-parameter MoE (16 of 896 experts active) claims third place on the Artificial Analysis intelligence index and tops six of seven frontend code categories on Arena.ai. API pricing is broadly comparable to DeepSeek V4 per token, though 17× more expensive than Kimi K2. via THE DECODER
  • Thinking Machines Lab releases Inkling, its first model. Mira Murati’s startup shipped a 975B-parameter, 41B-active multimodal MoE under Apache 2.0, trained on 45 trillion tokens of text, images, audio, and video. Benchmarkers place it as the leading US open-weight model on the Artificial Analysis index, though top Chinese open models still outscore it on several tasks. via Simon Willison
  • xAI files its first lawsuit over Grok-generated CSAM. After sustained reporting made it impossible to deny the model produces child sexual abuse material, xAI sued a user accused of generating the images, marking a significant shift in the company’s legal posture. via Ars Technica – AI
  • xAI open-sources Grok Build after a credential-exfiltration incident. The company’s coding CLI was found silently uploading full directories — including SSH keys and password databases — to Google Cloud. After public backlash, xAI released the 844,530-line Rust codebase under Apache 2.0 and committed to deleting all uploaded user data. via THE DECODER
  • EU formally orders Google to share search data and open Android to rival AI. European regulators issued the first binding decisions under the Digital Markets Act requiring Google to give competing AI assistants greater access to Android and provide rivals with search-index data. Both companies have one month to appeal. via The Verge – AI
  • OpenAI investigates GPT-5.6 Sol for unexpected file deletions in Codex. The company acknowledged reports that Codex had deleted user files in production, attributing the behaviour to running the model in full-access mode without sandboxing or auto-review enabled. via Simon Willison

Who shipped

Moonshot AI dominated the cycle with the Kimi K3 launch, claiming the open-weight frontier. Thinking Machines Lab (Mira Murati) posted its first open model, Inkling. Google rebranded NotebookLM as Gemini Notebook with an integrated cloud compute layer, added personal AI avatars to Google Vids via Gemini Omni, and extended AI Mode to connect with third-party apps; the company separately confirmed Gemini 3.5 Pro is months behind schedule on internal goals. xAI added Grok 4.3 to Amazon Bedrock for enterprise workloads while managing the two incidents above. NVIDIA released Nemotron 3 Embed 8B, which reached the top of the RTEB retrieval benchmark.

Open-source pulse

Kimi K3 (2.8T MoE, weights promised 27 July) and Inkling (975B Apache 2.0, available now) both pushed the open-weight frontier today, with benchmarkers placing Kimi K3 ahead of Opus 4.8 and Inkling as the top US open model. Gemma 4 received a silent update from Google fixing tool-calling bugs and truncated responses. The Grok Build Rust codebase was also released under Apache 2.0 following the credential-exfiltration incident.

Money, infra & hardware

Fireworks AI announced a $1.5B Series D at a $17.5B valuation after crossing $1B ARR; the company powers inference for Cursor and large enterprise deployments. Health-scan startup Neko Health raised $700M in a Series C led by Lightspeed to expand AI preventive-screening clinics to the United States. Energy IPOs are running at their fastest pace this century as investors seek data-centre power exposure.

Quiet corners

Anthropic posted no announcements and appeared only in benchmark comparisons. Meta was similarly absent; Llama 4’s earlier disappointing reception continues to define references to the lab. Mistral and Cohere were largely off the radar, the latter surfacing only in a university partnership note.

By the numbers

  • 429 stories today across ~30 sources
  • Most-mentioned model: Kimi
  • Most-mentioned lab: Moonshot AI
  • Notable absences: Anthropic, Meta, Mistral

Compiled by AI Feed’s editor from all 429 headlines published on 16 July 2026.