Extrapolating OpenAI's swarm scaling: Bel agents could run 1,024-hour tasks at 82% success
LeviTurk · x · 2026-10-17
Developer LeviTurk charted task success rate as a function of task horizon length, using OpenAI's claim (from "Research acceleration: The view inside OpenAI," Sep 2026) that internal time horizons double every 64 days, and extrapolated to their Bel model.
- Measured: success-with-nudges falls from 94% on <15m tasks to 67% at 64-128h; one-shot success drops to 13% beyond 128h.
- Extrapolated 10.6 months ahead with Bel: 82% success (with nudges) on 512-1024 hour tasks, 26% one-shot.
- His takeaway: ralph-looping Bel — restarting it continuously with occasional nudges — could let it keep working on long-horizon tasks indefinitely.
Related event: OpenAI Internal Data: Model Task Horizon Doubles Every 64 Days(2 posts)→
More from coding & agent
- Open-source coding agent OpenCode hits 214k GitHub stars with free DeepSeek V4.1 Flash built in — alex_verem · 2026-10-17
- AICre8 MCP Server lets agents build and deploy web projects in a sandbox — modelcontextprotocol · 2026-10-17
- Energetica MCP connector serves 40+ curated Argentine oil & gas tables — modelcontextprotocol · 2026-10-17
- Community ports DeepSelect to Hopper with CuTe DSL in TokenSpeed PR — zhyncs42 · 2026-10-17
- ComfyUI node turns illustrations into manga panels using Qwen-Image 2.1 Turbo — KnowledgeBig4364 · 2026-10-17
- Open-source CLI scans MCP configs for poisoned tool descriptions before agents run them — StanislavMZ · 2026-10-17