For small models, ipython is all you need: single-tool harness ties production-grade scaffolds on TB-Lite
yb2698 · x · 2026-10-14
The root post of the series: a single-tool harness can tie some complex production-grade harnesses on Terminal-Bench Lite — less scaffolding means better small agents. The benchmark is Terminal-Bench Lite from the openthoughts team (fast, same signal as TB 2.0, easier for small models), run with unsloth's Qwen3.5-9B GGUF weights on 7GB VRAM, reproducible on consumer laptops. Follow-ups cover tool-call distribution, prompt growth, stochasticity analysis, and a four-round ipython harness iteration.
More from coding & agent
- Startup launches cloud computers built for AI agents: 4 vCPUs in under a second — CurieuxExplorer · 2026-10-15
- User has Grok bot build its own email UI, and it just works — elonmusk · 2026-10-15
- YC F26's Papaya launches to auto-fix production AI agents via PRs — ycombinator · 2026-10-15
- Eval-driven development for RAG: keep eval cases in Git, never regress — Al_Grigor · 2026-10-15
- LlamaIndex launches OpenDocRouter with ParseBench benchmark comparisons — llama_index · 2026-10-15
- User reverse-engineers Grok's X news curation prompt and turns it into a reusable agent skill — dotey · 2026-10-15