jul 10, 2026
5 links from the engineering internet.
zed 1.10.2 wires up gpt-5.6 sol and terra for chatgpt subscribers
zed's assistant can now use openai's new gpt-5.6 sol and terra models for users on a chatgpt subscription. the lighter luna tier stays gated, since openai has not opened it to third-party clients yet.
llama.cpp b9949 adds cluster-parallel flash-attention decode for adreno gpus
the opencl backend gains cluster-parallel decode for flash attention on qualcomm adreno gpus, speeding up token generation for on-device inference on android and other adreno hardware.
litellm v1.93.0-dev.3 adds passthrough and oauth-delegate mcp auth types
the dev build exposes true_passthrough and oauth_delegate auth types for mcp servers behind a no-auth warning, and makes bedrock guardrails honor disable_exception_on_block by raising a modify-response exception instead of erroring.
n8n 2.30.3 keeps the ai assistant thread intact on editor hand-off
the pre-release fixes a bug where the ai assistant's conversation thread was lost when the editor handed off between views, so in-progress help now survives the transition.
next.js 16.3 canary teaches turbopack to compile pages-router service workers
the canary makes turbopack compile service workers registered from pages-router pages and emit them to /_next/static/, closing a gap where pages-router service workers were not bundled correctly.