Thinking Machines releases Inkling 2 Mini, a looped transformer topping sub-100B open-weights coding benchmarks
SonglinYang4 · x · 2026-11-21
Thinking Machines Lab released Inkling 2 Mini, an open-weights looped transformer that runs the model's middle layers three times per token, letting it reason longer without writing its reasoning down.
Key facts
- MoE with 26.4B total params, 3.4B active, 5.4B weight activations per token counting loops
- Leads all open-weights models under 100B total params on DeepSWE v1.1 (48.2%) and Vals Index v2.2 (46.5%); runs on a laptop and faster than similarly performing models
Safety findings (via a variable-loop research variant also released)
- No-CoT time horizon improved from 16s at one pass to 37s at three
- Serial-reasoning benchmark performance kept improving out to ten loops—twice the training count
- Looping makes reasoning implicit, directly affecting chain-of-thought monitorability; weights are open so the AI safety community can study how loops affect CoT oversight
More from AGI Musings
- Two Riemann hypothesis proofs announced same day: one with Claude, one "all human, no AI" — Dr_Atoosa · 2026-11-21
- Mathematicians will study Lean proofs like N-body simulation results, researcher predicts — mjuric · 2026-11-21
- METR: AI R&D capabilities are improving fast — and getting harder to measure — dfrsrchtwts · 2026-11-21
- Taking AI Seriously: Borretti traces how programmer AI advice keeps retreating — zetalyrae · 2026-11-21
- Sundar Pichai memes Euler and Gauss gawking at a 'Riemann Hypothesis Proved' headline — sundarpichai · 2026-11-21
- Blogger argues starving AI of capital is the real way to stop the singularity — JensHonack · 2026-11-21