AI Coding Live News

On 2026-10-01 Earendil shipped Pi 1.0 (MIT): a hardened minimal coding-agent harness. Built-ins: Codemode (QuickJS sandbox for model-written JS calling tools / Jev / image models) + MCP, virtual-model routing, deferred tools, Anthropic cache warming, mid-conversation system edits. At ingest: 111793★; HN 1645 pts / 573 comments. Experimental Pi Durable shipped the same day.

Mastra published @mastra/[email protected] on 2026-09-24 with ModelSelectionProcessor for classifier-backed per-request model routing (#24865), closeOnSuspend on durable agent streams so AG-UI-style loops can finish on tool suspend, plus createClassifierScorer(), optional tool titles, and trace aggregate dimension/measure helpers with tag predicates. Docs warn routing is not always cheaper because models do not share prompt caches.

@DanKornas highlighted TokenRhythm/opensquilla, a token-efficient microkernel AI agent for builders. SquillaRouter scores each turn locally on length, language, code, keywords, and semantic embeddings, then routes across four tiers to the cheapest capable model; the registry targets 20+ LLM backends with primary-plus-fallback. Fifteen bundled skills load on demand, and the agent can act as an MCP client or server. Local Markdown notes add SQLite full-text search and sqlite-vec semantic recall, with three sandbox policy tiers plus human-in-the-loop approvals for sensitive tools. Apache 2.0.

Zhe (@iamzhe) shipped a minirouter.sh update: minirouter/auto for task-based model routing, Google/GitHub/passkey login, and a fix for Codex users with Qwen models.

Cline reported that Zhipu GLM-5.3 at max effort outperforms OpenAI GPT-5.6 Sol at max effort on the newly released Terminal-Bench 4.0. The post quotes Zhipu lead Zixuan Li's announcement that the benchmark itself is iterating fast enough to keep up with model releases. Cline already hosts glm-5.3 in-product: install with npm i -g cline, run /model, and select glm-5.3. ClinePass is offered as roughly 5x discounted access versus calling the direct API, continuing Cline's push to make open-weight agent models the default path rather than a hobby sidecar. For teams running terminal-native coding agents, the result matters because Terminal-Bench measures long-horizon shell work rather than single-file completion. An open-weight win against Sol on that surface changes both model routing and budget planning, especially if ClinePass pricing holds under production traffic.

Replit launched Intelligent Model Routing for Replit Agent, automatically picking the best model per task with an auto toggle, cutting development costs by up to 65% while maintaining output quality.

Cloudflare introduced a 50% promotional discount for OpenAI's GPT-5.6 Sol via AI Gateway unified billing, combining edge caching, observability, and cost-effective multi-model routing.

Cursor launched Cursor Router, an intelligent model dispatch engine. In Auto mode, users configure goals for Intelligence, Balance, or Cost. The router sends complex tasks to frontier models and routine edits to efficient ones, cutting cost per commit by ~60% with zero quality degradation.