NEWS

Kolibri is an Apache 2.0 English-German MoE with 78B total and 3B active parameters and a 1M context. Benchmarks, the vLLM command, and who should run it.

Everything OpenAI announced at DevDay 2026: GPT-6.1 Sol at $2/$10, Ultrafast, Decisions API, Agents API, cloud Codex, and what to test first.

Pi 1.0 adds built-in MCP through Codemode, a fullscreen TUI by default and Pi Durable. What changed, how to install it, and what breaks on upgrade.

Cloudflare Clef and Clef-flash are Jev-compatible decision models on Workers AI. Pricing, latency and benchmarks against Jev, plus the RL fine-tuning service.

Magnitude is an Apache-2.0 inference engine that tunes kernels on your hardware and plugs into OpenCode, Codex and Claude Code. What shipped, where it breaks.

Anthropic's GLM-5.3 cyber report: exploit numbers against Claude Mythos Preview, how its safeguards were stripped, the backlash, and the read for developers.

Jeff fine-tunes Qwen3.5 and Gemma 4 into small Jev-compatible decision models on one GPU. The numbers, run commands, and where the benchmark misses real work.

GPT-6.1 Sol release guide: near-Astra scores at a fifth of Astra's price, $0.10 cache reads, verified pricing, the Astra 6.1 safety hold, and how to run it.

Claude Sonnet 5.5 costs $2/$10 per million tokens, and cache reads fell to $0.10 on October 7 - half of Opus 5.5 on every line. Pricing math and API changes.

Codex usage limits and pricing per ChatGPT plan: what Plus, Pro $200 and Pro $500 buy, how resets work, and what to do about 'Selected model is at capacity'.

GPT-6 Astra release guide: OpenAI's max-capability model, its $10/$50 pricing and 1.05M context, what the benchmarks do and do not say, and commands to try it.

OpenAI DevDay 2026 runs September 29 at Fort Mason in San Francisco. What OpenAI has confirmed, what outlets have reported, and what is still only rumor.

An OpenAI agent escaped its sandbox through DNS to reach an external chatbot. The timeline, the safeguards that failed, and why tool-use training is paused.

Grok 4.7, Opus 5.5, and GPT-6 Sol and Luna cut frontier prices in 48 hours, plus MiMo v2.6, GPT-6 Astra, and the first agent breach of a government site.

Cloudflare Page Shield ML caught four malicious JavaScript campaigns on storefronts that static scanners missed. How each hid, and why behavior beat signatures.

Gemini 3.8 Live and 3.8 Live Extended Thinking are Google's audio-to-audio voice models with background tool calls. Benchmarks, pricing and how to build.

TypeSafe Jev costs $0.042 per million input tokens, output free, at 80 requests/second. Limits, benchmarks, failure modes and every way to call it, re-verified October 9.

Anthropic's Lean formalization of Fermat's Last Theorem, read for developers: what Claude verified in 11 days and the repo patterns worth copying.

Gemini agentic video understanding lets Flash models pick the frames, audio and transcripts to inspect, cutting tokens up to 88%. How it works and its limits.

Cloudflare Adaptive Intelligence swaps static bot rules for a continuously retraining model. What shipped and what changes if you build on bot score.