Newsletter Wednesday 4 November 2026
Daily update on the AI tools we follow. Headlines come from public sources (X, GitHub, Hacker News); the Dutch commentary and summary are machine‑written. Each headline links to our own news page on that topic; the source link is shown beside it.
In brief
Anthropic has released FlashProve, an open‑source zkSNARK tool that can verify the provenance of LLM output. Liquid AI launched the LFM3‑1.7B model with 1.7 billion parameters and a 256K context window for local agents and edge deployment. Additionally, the native multimodal model LongCat‑2.5 has appeared. In the past month, a total of ten new open LLM architectures were released.
Claude Code & agents
- Anthropic releases FlashProve: open-source zkSNARKs for provable LLM inference. Prove a response came from a specific model (without reveali · source: X · 2026-11-03
Anthropic has released FlashProve, an open‑source zkSNARK tool to verify LLM output, helping indie developers prove model provenance.
Models & pricing
- Liquid AI releases LFM3-1.7B, a 1.7B tri-hybrid model with a 256K context window, designed for local agents, tool calling, and edge deployme · source: X · 2026-11-02
Liquid AI has launched LFM3‑1.7B, a 1.7 billion‑parameter model with a 256 KB context window, suitable for local agents and edge deployment, enabling indie developers to run powerful models on‑device. - The most useful line in this week's model releases was about failures, not benchmarks. LongCat-2.5, released on Nov 1, is a native multimoda · source: X · 2026-11-02
LongCat‑2.5, released on 1 November, is a native multimodal model that offers indie developers an alternative for multimodal applications.
Repos, releases & updates
- Catching up on the 10 (!) open LLM architecture releases from the past month. Quite a lot happened in October. I guess we can call it "Open · source: X · 2026-11-03
Ten open LLM architectures have appeared in the past month, giving indie developers more choices for open‑source models.