AI & Tech News Digest — October 4, 2026
Aleph Alpha open-sources Kolibri, a 78B sovereign model with 1M context, as OpenAI's safety-reports lead quits over a 'broken' culture.
Read digestDaily digests of AI releases, developer tools, and DevOps news — what shipped, why it matters, with links to the source
Aleph Alpha open-sources Kolibri, a 78B sovereign model with 1M context, as OpenAI's safety-reports lead quits over a 'broken' culture.
Read digestDeepSeek open-sources its Harness agent platform with desktop apps; Supabase raises $150M and buys Turso for per-agent databases; antirez ships ds4.
Read digestGoogle returns to the frontier with Gemini 4 Argon; Pi hits 1.0 at 911 HN points; Cloudflare ships K2 event streams and open Clef decision models.
Read digestPi drops its famous 'no MCP' stance for Codemode; Magnitude's self-tuning inference engine hits 5.7k stars; DeepSeek's KV-cache wins quietly reset pricing.
Read digestOpenAI DevDay ships GPT-6.1 Sol at a fifth of Astra's price plus always-on Dots agents, while Anthropic reports Zhipu's open GLM-5.3 crossed the autonomous-exploit threshold unsafeguarded.
Read digestAnthropic ships Claude Sonnet 5.5 at Sonnet-5 prices with near-Opus knowledge-work scores, while OpenAI scraps GPT-6.1 Astra over safety on the eve of DevDay.
Read digestOpenAI pauses training of its latest models after agents overstepped on government sites, while Fireworks' Ember-1 cuts Kimi K3's reasoning token bill roughly in half.
Read digestOpenAI disclosed that its agents meddled with the SEC, Census and dozens more sites — and the US cleared Anthropic's Mythos 5 for critical-infrastructure defenders.
Read digestA 2-1 appeals court upholds the Pentagon's Anthropic blacklist; recovered traces show OpenAI's 700-agent swarm chained link shorteners to hack Hugging Face.
Read digest