
Moonshot unveils 2.8T Kimi K3 with 1M-token context as open-source models surge — Codex tops 9M users and Suno data breach raises alarms
Share your love
Hello from Aurora and Isabelle
Salute: Hello listeners — this is AI Tech News Today.
Date: 2026-07-17T11:31:01.000Z
We have a quick roundup of the biggest A‑I headlines you need to know, spoken clearly so your commute does not panic.
1) Moonshot AI unveils Kimi K3
What’s new: Moonshot AI revealed Kimi K3, a new frontier model.
- Size: ~2.8 trillion parameters
- Context window: 1,000,000 token context
- Capabilities: Native multimodal support for text, images, and more
- Focus: Architecture optimized for agent‑style coding and efficiency
- Availability: Team plans to open the model weights on July 27
If you like big contexts and local tinkering, mark that date.
2) Codex passes nine million users
Snapshot: Codex and ChatGP‑T Work surged past 9 million users in the week after the release of GPT 5.6.
- Demand was so high teams issued repeated weekly limit resets to keep people running.
- Rapid growth means more people experimenting with A‑I–assisted coding — and more pressure on infrastructure and safety guardrails.
3) Thinking Machines releases Inkling
What: Startup Thinking Machines shipped its first model, Inkling, and published full weights.
- Multimodal: Reasons across text, image, and audio
- Tunable: Offers configurable cost/performance tradeoffs
- Performance: Reportedly among the strongest open models on audio benchmarks
If you build audio or multimodal apps, this is worth benchmarking.
4) Xai open‑sources Grok Build
What: Xai made Grok Build fully open source.
- Included: Complete agent loop, code editing & execution tools, a terminal UI, and an extensible plugin/subagent system
- Design goal: Local‑first with user‑controlled inference — useful for developers who want agentic coding without sending everything to a remote service
5) Suno‑AI training data breach
What happened: Music A‑I maker Suno reported a breach exposing details of its training dataset.
- Exposed material: Hundreds of thousands of hours of music and plans to ingest a million hours of podcasts
- Implications: Raises questions about consent, licensing, and how companies protect the raw feeds that teach their models
Closing — stay informed
That is your quick brief. We will keep following releases, open source moves, and security wake‑up calls in A‑I. Stay curious, stay skeptical, and we will see you next time with more news.














