#open-weight_

Writing

EU AI Act for developers: what you have to do when you build on AI modelsWhat the EU AI Act asks of apps built on API or open-weight models: provider or deployer, Article 50 labels and watermarks, high-risk rules, deadlines, fines.2026-10-08
Aleph Alpha Kolibri: benchmarks, hardware and how to run itKolibri is Aleph Alpha's open-weight German-English model: 78B parameters, 3.5B active, Apache 2.0. How it scores, what GPUs it needs and how to run it.2026-10-07

Daily

Haiku 5.5 and four-cent decision models: the AI price war reaches the bottom tierLast week three labs launched mid-tier models at $2/$10 list prices; this week the fight moved below that, to Claude Haiku 5.5 at GPT-6 Luna's price, decision models at about four cents per million tokens from four companies, and routers that pick the model for you. Also, agent containment became a platform feature built into Windows and Copilot and incident disclosure a White House demand, Western labs pitched open weights on sovereignty and fewer refusals, and AI bug-finding outran the people who fix the bugs.Saturday, 10 October 2026 · Saturday edition
Google launches the Gemini agent, one work agent that also runs Claude modelsGoogle's Gemini agent puts chat, long-running tasks, code and coworker agents with their own email addresses behind one prompt box and API, orchestrating Gemini and Claude models. Arena's new Alignment Index finds agents claim unfinished debugging work is done in 48% of sessions, Anthropic launches OSS Scanner, free, opt-in, unreviewed security scans for open-source maintainers, and rewrites its usage policy for hardware, surveillance and agent work, and JetBrains releases Mellum2.1, an open 12B coding model with 2.5B active parameters.Thursday, 8 October 2026
OpenAI publishes 722 AI-written math papers, claiming proofs of the quasi-Riemann hypothesis and Unique GamesOpenAI releases 722 manuscripts from an unreleased internal model, many checked in Lean, claiming results including the quasi-Riemann hypothesis and Khot's Unique Games Conjecture, Mistral opens a preview of Large 4, a 1T-parameter open-weight model with weights due this month, Anthropic lifts Claude's cyber blocks for verified defenders in three tiers, OpenAI launches a Decisions API for fast classification at $0.10 per million input tokens, Google releases EmbeddingGemma 2 for on-device multimodal search, and Polars 2.0 makes streaming the default.Tuesday, 6 October 2026
OpenAI starts watermarking ChatGPT text in the EU and lets API users opt inOpenAI will watermark ChatGPT and Codex text in the EU and offers opt-in watermarking in the API, GitHub releases ReviewBench, an open benchmark for AI code review, Together AI's Together Link puts open models inside Claude Code and Codex, Reflection previews Beam, a 501B open-weight model, Cohere ships North 2, and the Pentagon says it has stopped using Claude.Monday, 5 October 2026
Aleph Alpha releases Kolibri, an Apache-licensed German-English modelAleph Alpha's Kolibri is a free open-weight German-English model, 78B parameters with about 3.5B active, a 1M-token context and Apache 2.0 weights on Hugging Face, built for on-premises use in Europe. Also, a Rejetto HFS flaw found by Anthropic's Mythos was being exploited a day after it was disclosed, and Microsoft's ThinkingBox benchmark shows how few agent tasks any model gets right 20 times in a row.Saturday, 3 October 2026