ThinkyMachines' Inkling 2 Small trains models on its own: $426 of compute doubles Qwen score in 4.5 hours
Scobleizer · x · 2026-11-06
ThinkyMachines launched Inkling 2 Small, scoring 31 on the AA Intelligence Index — a big jump over its predecessor. It's available now for fine-tuning and inference on Tinker, with full weights coming.
The launch demo let the model do an ML engineer's afternoon of work: given Qwen3.5-4B, it wrote its own evaluation, ran its own training experiments, and after 4.5 hours and $426 of compute more than doubled Qwen's forecasting score.
Weights are due under Apache 2.0 next week, so anyone with the hardware can run the loop and keep what it builds.
More from coding & agent
- alvelda: treat agent memory corpus as a trainable model component, not just a data format — alvelda · 2026-11-06
- SSH turns any VPS into an agent node: Codex, Claude Code, Kimi all know how to delegate remotely — huangyun_122 · 2026-11-06
- Indie dev: the key to better agent workflows is ceding more 'safety margin' to AI — yihui_indie · 2026-11-06
- Managing all your AI agents in one place: a screenshot tour of approval-gated computer use — RileyRalmuto · 2026-11-06
- Managing 30,000 Notes with Local Hybrid Search: qmd, MCP Servers and AI Assistant — dSebastien · 2026-11-06
- Skillware framework claims 97.3% cost cut as 3B-8B local models beat frontier LLMs — RossPeili · 2026-11-06