By use case · Reasoning
Best local reasoning models
These models think step by step before they answer, which helps on math and logic and costs more time per reply. 18 open reasoning models are ranked by quality below. The strongest is DeepSeek R1 (~383.7 GB at Q4_K_M); the lightest is LFM2.5 1.2B Thinking (~1.8 GB).
Models that think step by step before answering, for math, logic and planning.
- 1 ~383.7 GBDeepSeek R1671B MoE · runs on 0/40 devices
- 2 ~384.1 GBDeepSeek-R1-0528671B MoE · runs on 0/40 devices
- 3 NE ~341 GBNemotron 3 Ultra 550B-A55B550B MoE · runs on 0/40 devices
- 4 NE ~80.3 GBNemotron 3 Super 120B-A12B120B MoE · runs on 4/40 devices
- 5 NE ~30.6 GBLlama-3.3-Nemotron-Super-49B-v149B · runs on 8/40 devices
- 6 ~22.1 GBDeepSeek-R1-Distill-Qwen 32B32B · runs on 12/40 devices
- 7 NE ~25.1 GBNemotron 3 Nano 30B-A3B30B MoE · runs on 9/40 devices
- 8 NE ~25.1 GBNemotron Cascade 2 30B-A3B30B MoE · runs on 9/40 devices
- 9 ~16 GBMagistral Small24B · runs on 15/40 devices
- 10 ~10.7 GBDeepSeek-R1-Distill-Qwen 14B14B · runs on 24/40 devices
- 11 ~10.1 GBPhi-4-reasoning14B · runs on 29/40 devices
- 12 NE ~7.6 GBNemotron Nano 9B v29B · runs on 33/40 devices
- 13 ~6.2 GBDeepSeek-R1-0528-Qwen3-8B8.19B · runs on 33/40 devices
- 14 ~6.4 GBDeepSeek-R1-Distill-Llama 8B8B · runs on 33/40 devices
- 15 ~6.1 GBDeepSeek-R1-Distill-Qwen 7B7B · runs on 33/40 devices
- 16 NE ~3.9 GBNemotron 3 Nano 4B4B · runs on 40/40 devices
- 17 ~3.6 GBPhi-4-mini-reasoning3.8B · runs on 40/40 devices
- 18 LF ~1.8 GBLFM2.5 1.2B Thinking1.17B · runs on 40/40 devices
Tagged by each model's stated purpose. The memory figure is what it needs at Q4_K_M, and the device beside it is the lightest tracked machine that fits it. "Runs on N devices" counts the 40 tracked devices that fit it at Q4_K_M.
FAQ
What is the best local reasoning model?
DeepSeek R1 is the highest-ranked, but it needs ~383.7 GB and no tracked device runs it locally. The strongest one you can actually run is Nemotron 3 Super 120B-A12B. Pick by what fits your memory using the list above.
What is the smallest reasoning model that runs on a laptop?
LFM2.5 1.2B Thinking is the lightest, at ~1.8 GB at Q4_K_M, so any device with 8 GB or more can load it.
How were these reasoning models chosen?
They are open-weight models whose own design targets reasoning (by name, family or model card). Memory figures are computed at Q4_K_M and sourced; see the methodology page.
Memory is computed at Q4_K_M, validated 2026-08-03. See methodology.