[AINews] Pi 1.0, Pi Durable, and AIE NYC

Earendil's Pi 1.0 and Pi Durable both hit the front page of Hacker News, according to the AINews roundup. Pi 1.0 adds Codemode with native support for MCP, Jev and image models, extension support for virtual models, deferred tool loading, cache warming for Anthropic models, mid-conversation system messages described as transcript-aware prompt and tool changes, a new TUI theme, and full-screen mode by default.

Pi Durable ports Pi to TypeScript and externalizes all stateful components of Pi. Listed features include crash survival, where every step is recorded as a checkpointed task so agents and subagents automatically resume from their last exact state after a process failure or restart; portability to any JavaScript runtime such as Node, Bun or Cloudflare, with pluggable storage backends (Memory, SQLite, JSONL) and flexible remote or local execution environments; concurrency across multiple parallel, branching conversations; extensibility through installable Extensions bundling custom system prompts, tools, hooks and durable tasks; automatic background compaction of older messages for context management; multiplayer and state sync with application state stored in documents alongside conversation transcripts; and hot-swapping of tool and extension code while an agent runs.

The roundup also covers other launches. Google announced Gemini 4 Argon, with contributors citing revised pretraining mixtures, long-horizon post-training data, and internal applications; Google says new Gemini revisions undergo weeks of testing by thousands of internal software engineers. A circulated Bloomberg report attributed coding weaknesses to anonymous insiders, and a subsequent post reported a senior DeepMind engineer rejecting that account; the practical coding-quality dispute is described as unresolved.

OpenAI's GPT-6.1 Sol is framed as an efficiency update. Sam Altman called it the company's fastest-growing model and said serving performance improved after launch-time load problems. Artificial Analysis reports $0.72 per Intelligence Index task at maximum effort, versus $1.04 for GPT-6 Sol and $3.26 for Astra, attributing the gain to fewer turns and cheaper cache reads. OpenAI also corrected image encoding for Luna and Sol.

Upstage's proprietary Solar Mini 4 reports 35B total/3B active parameters, a 1M-token context window and 262K maximum output, with weights not released. Black Forest Labs launched FLUX 3 Image with native generation up to 4K, up to ten reference images, bounding-box layout control and multi-turn editing, plus a temporary 50% API discount through October 8. Tavus introduced Griffin, a video-to-video interaction model.