Ant Group's inclusionAI published MIT-licensed weights for Ling-3.1-flash on Hugging Face—a hybrid-linear MoE with 560B total / 25B active parameters and a 1M-token context, with day-0 SGLang and vLLM recipes. The card also teases LingCode, a long-horizon coding harness built on the open-source Pi agent that they may open-source later.
BAND launched BAND for Developers on Oct 15: a macOS/Windows/Linux desktop app (plus CLI) that gives a visual view of projects, agents, ownership, blockers, and progress so agents across different harnesses and machines can share context and hand off work without you ferrying every task…
OpenAI's openai agents python v0.24.0 (Oct 15) exposes MCP annotations on tools converted from MCP servers and adds run wide tool exception formatting via RunConfig.tool error formatter . It also batches OpenAI conversation session writes within API limits and defers automatic compaction while approvals are pending…
Figma opened Specs in open beta on Oct 15—multiplayer Markdown containers on the Design and FigJam canvas that embed frames, diagrams, code layers, and Code Connect. Paste a Spec link into a coding agent on Figma’s MCP server and it gets that full context; agents can also update the Spec as the build changes…
AWS Labs' MIT 0 AI DLC 2.11 installs a native aidlc command that runs one harness neutral workflow engine inside Claude Code, Cursor, Codex CLI, opencode, Kiro, and Copilot, picking a workflow from your /aidlc request and stopping at human approval gates. Set it up per project with aidlc config harness <name and check it with aidlc doctor…
StepFun released open weights for Step 5 Preview, a 600B total / 27B active sparse MoE pitched at repo level software engineering and long horizon tool use, with up to 1M tokens of context and output. FP8 runs on one 4x B300 or 8x H200 node via day 0 vLLM and SGLang images, under the Apache style StepFun Model License 1.0 with extra commercial terms…
Spotted something we missed? Start a thread.