The industry dropped the ball on strong, tool-calling, non-reasoning small LLMs
vboykis · x · 2026-09-23
Yoav Goldberg argues there really is a market for strong, instruct-tuned, tool-calling, non-reasoning, non-conversational, small-context LLMs, but the industry dropped the ball on them and hopefully we'll see a revival. Vboykis adds that many tasks don't need a reasoning LLM — being able to define a task in text and execute it is enough — and for recurring tasks, one could go further by collecting examples and tuning a tailored predictor.
More from Models
- Cognition floods Devin with GPT-6 models and cuts task costs 61%, gives away 50 Max plans — EricBuess · 2026-09-23
- GPT-6 Astra beats Claude Opus 5.5 31s in LLM-driven robot sumo sim — DJiafei · 2026-09-23
- User's article keeps tripping Opus 5.5 safety monitors — 'I guess I wrote an infohazard' — scaling01 · 2026-09-23
- CliffCompaction installs in two commands; DeepSeek v4.1 compacts best, says Dettmers — Tim_Dettmers · 2026-09-23
- Dettmers: CliffCompaction works much better with full thinking traces — Tim_Dettmers · 2026-09-23
- With CliffCompaction, open-weight models beat closed ones in long-horizon sessions — Tim_Dettmers · 2026-09-23