Thinking Machines launches Inkling 2 Small: 594B MoE, Apache 2.0 next week
echen · x · 2026-11-06
Thinking Machines Lab introduces Inkling 2 Small, its second-generation model: a 594B-parameter (17B active) MoE with image input and variable effort, scoring 31 on the Artificial Analysis Intelligence Index. Key facts:
- Availability: initially closed-weights via Tinker (playground, fine-tuning, sampling); full weights to be released under Apache 2.0 next week after external safety testing and red-teaming.
- Benchmarks: model card cites 6 Surge benchmarks — sudo L7, DAYJOB: Finance, Chartography, ComplexConstraints, EnterpriseBench: CoreCraft, GDP.pdf. On sudo L7 it scores 14.1 at $0.34/task, within 2 points of models 8-13x more expensive.
- Agentic demos: iteratively builds websites/games/3D designs with visual feedback; a Three.js city simulator generated entirely procedurally; a from-scratch Python PC emulator that boots real Linux in the browser.
- Autoresearch: autonomously trained Qwen3.5-4B on Tinker for time-series forecasting in 4.5 hours and $426 of spend, more than doubling the forecasting score.
- Improvements attributed to more efficient pre-training and a large unified RL run.
Related event: Thinking Machines Releases Inkling 2 Small Open-Weight Model(42 posts)→
More from coding & agent
- AI revives Facebook's abandoned Prepack, speeding up React renderToString 5.4x — Vjeux · 2026-11-06
- If You're Compute-Constrained, Don't Burn Agent-Compute on Social Agents — nptacek · 2026-11-06
- AI Engineer SH talk: models see the world through their action space, not ours — vincent_koc · 2026-11-06
- teortaxesTex: Singularity still needs bigger models; public AGI easy to dunk on — teortaxesTex · 2026-11-06
- Late to a talk, his AI agent tapped the live transcript and took notes for him — vincent_koc · 2026-11-06
- VoidZero October Update: Rust-Based TypeScript 8x Faster, SDK Goes Open Source — cnakazawa · 2026-11-06