OpenAI ships compaction in Responses API; devs debate if it kills differentiation
zeeg · x · 2026-10-06
A thread on OpenAI shipping compaction (context compression) built into the Responses API endpoint. @pvncher argues compaction is a core model capability that is very hard to separate out during training, and developers can build a custom harness on Responses API and use the official compaction. @zeeg counters that this exposes how limited the tools vendors give customers are — leaving no room to build a competitive product or vary the implementation — and doubts this approach survives even if he understands why providers hide reasoning.
Related event: OpenAI's Built-in Compaction Sparks Vendor Lock-in Debate(10 posts)→
More from coding & agent
- HeyGen ships GPT-Live language tutor: full-duplex speech model drives an avatar teacher — HeyGen · 2026-10-06
- Modly makes llama.cpp its default agent engine for local 3D mesh generation — Lightnig125 · 2026-10-06
- Dev sketches a subvocal "exocortex": silent speech prompt sent to a workstation LLM agent — cephaloform · 2026-10-06
- Building a Context Graph: 6 Months of Learnings Toward Self-Improving Software — whoiskatrin · 2026-10-06
- Meta reportedly scales back internal Claude use, pushing employees to MetaCode & Muse Code — Polymarket · 2026-10-06
- TanStack Launches MCP Server for AI-Assisted Docs Search and Scaffolding — modelcontextprotocol · 2026-10-06