Running a coding builder agent on GTX 1080 Ti 11GB: model candidates tested
NegotiationNo1504 · reddit · 2026-11-08
A Redditor seeks a local LLM to act as a builder agent on GTX 1080 Ti 11GB + 32GB RAM (Debian, llama.cpp/Vulkan since CUDA is unstable), paired with an API planner model to cut costs, targeting 100k+ context. Tested so far: Qwen3.6 35B A3B UD Q4KM (7–10 t/s, under-tested), Qwen3.5-9B MTP (25–30 t/s but inadequate as builder), Sharp Spark X2.5 4B (overthinks, unusable), X2.6 4B (fine for simple tasks only). Asks for working model/quant/harness combos at Q3/Q4 for real agentic coding.
More from coding & agent
- Anthropic's guide: turn Claude into a work assistant with Skills and MCP — Arindam_1729 · 2026-11-08
- paper-toss: open-source toy where deleting a file crumples it into a paper ball tossed into a bin — prasenx · 2026-11-08
- Dev benchmarks tiny LLMs on agent tasks: 9B passes 2/5, stepping up to 12B — draginol · 2026-11-08
- Engineer proposes CI/CD could become a handoff to a deployment agent — andreisavu · 2026-11-08
- Dev launches Local AI benchmark scoring models by hardware on agent tasks and speed — draginol · 2026-11-08
- Personal AI dot can now read inbox sign-in codes to log in on your behalf — Dimillian · 2026-11-08