llmrun

Best Local Reasoning LLMs in 2026

Best Local LLM for Reasoning by VRAM

  1. 8 GB

    RTX 5060

    8 GB tier

    Best pick
    Qwen3.5 9B97

    Q4_K_M · 6.6 GB · 46 tok/s

    Alternative
    Llama 3.1 8B Instruct69

    Q4_K_M · 7.9 GB · 55 tok/s

  2. 12 GB

    RTX 5070

    12 GB tier

    Best pick
    Gemma 4 26B A4B IT102

    IQ3_S · 12 GB · 162 tok/s

    Alternative
    Qwen3.5 9B97

    Q4_K_M · 6.6 GB · 69 tok/s

  3. 16 GB

    RTX 5080

    16 GB tier

    Best pick
    Qwen3.8 27B124

    Q3_K_L · 15.4 GB · 42 tok/s

    Alternative
    Gemma 4 26B A4B IT102

    Q4_K_S · 15.5 GB · 173 tok/s

  4. 24–48+ GB

    RTX 4090

    24 GB tier48 GB tier

    Best pick
    Qwen3.8 27B124

    Q4_K_M · 17.9 GB · 38 tok/s

    Alternative
    Gemma 4 31B IT110

    Q4_K_M · 21.7 GB · 32 tok/s

Highest llmrun Reasoning Score among official releases that fit each memory size at a Q3-or-better quant; speed is decode tok/s on that size's reference GPU. Full reasoning ranking

Reasoning models think step-by-step before answering, excelling at math, logic, and multi-step problems. Thanks to distillation, you no longer need a data center to run them: compact distills of DeepSeek R1 and models like QwQ-32B bring chain-of-thought reasoning to a single consumer GPU. Below are the open-weight reasoning models you can run locally, ranked by their llmrun Score for reasoning — pick one to see the exact hardware that runs it.

168 Reasoning Models You Can Run Locally

Top 15 by llmrun Reasoning Score, official releases that fit 48 GB or less

  1. 1

    Qwen3.8 27BAlibaba · 27.8B

    Alibaba's Qwen3.8 27B is a dense, vision-capable model with around 28 billion parameters, continuing the Qwen line into its 3.8 release. It can process images alongside text prompts, supporting tasks like visual question answering and image-grounded chat in addition to…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    124

    Q3_K_L · 15.4 GB

    42 tok/s

  2. 2

    Qwen3.5 27BAlibaba · 27.8B

    Qwen3.5 27B is a 27.8-billion-parameter dense model from Alibaba's Qwen team, built for both text and image input using the same hybrid linear/full-attention architecture as the rest of the Qwen3.5 family. As a vision-capable model it can describe and reason…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    111

    Q3_K_L · 15.4 GB

    42 tok/s

  3. 3

    Gemma 4 31B ITGoogle · 31.3B

    Gemma 4 31B IT is Google's 31-billion-parameter instruction-tuned model in the Gemma 4 lineup, built to handle both text and image input in a single pass. As a vision-capable model, it can describe, compare, or reason about images alongside written…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    110

    Q4_K_M · 21.7 GB

    32 tok/s

  4. 4

    Qwen3.5 35B A3BAlibaba · 36.0B

    Qwen3.5 35B A3B is Alibaba's mixture-of-experts vision-language model in the Qwen3.5 line, totaling 36 billion parameters with about 3 billion active per token (the A3B in its name). It accepts text, images, and video as input, using a hybrid linear-attention…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    107

    Q4_K_M · 22 GB

    151 tok/s

  5. 5

    Qwen3.6 27BAlibaba · 27.8B

    Released a few months before its Qwen3.8 sibling, Qwen3.6 27B is Alibaba's dense, roughly 28-billion-parameter model from the Qwen 3.6 generation. Vision input is supported alongside text, so the model can handle image-based prompts as well as standard chat, coding…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    105

    Q3_K_L · 15.4 GB

    42 tok/s

  6. 6

    Qwen3.6 35B A3BAlibaba · 36.0B

    Qwen3.6 35B A3B pairs a mixture-of-experts design with roughly 35 billion total parameters, only about 3 billion of which are active per token. That small active footprint keeps generation speed close to a 3B-class dense model, even though the full…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    104

    IQ3_S · 15.7 GB

    181 tok/s

  7. 7

    Gemma 4 26B A4B ITGoogle · 25.8B

    Gemma 4 26B A4B IT applies a mixture-of-experts design to Google's Gemma 4 family, with roughly 26 billion total parameters but only about 4 billion active for any given token. That active-parameter count is what determines inference speed, so despite…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    102

    IQ3_S · 12 GB

    162 tok/s

  8. 8

    Qwen3.5 9BAlibaba · 9.7B

    Qwen3.5-9B is Alibaba's dense 9-billion-parameter language model with a built-in vision encoder, part of the Qwen3.5 generation. This is the post-trained, instruction-tuned release, combining Gated DeltaNet and gated-attention layers for efficient long-context inference, and it can take images alongside text…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    97

    Q4_K_M · 6.6 GB

    46 tok/s

  9. 9

    Qwen3 30B A3B Thinking 2507Alibaba · 30.5B

    Qwen3 30B A3B Thinking 2507 is the reasoning-focused variant of Alibaba's 30-billion-parameter mixture-of-experts model, updated in July 2025. Like its instruct sibling, it activates only about 3 billion parameters per token, keeping resource demands low while enabling multi-step reasoning and…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    94

    Q3_K_M · 16 GB

    188 tok/s

  10. 10

    Qwen3 30B A3B Instruct 2507Alibaba · 30.5B

    Qwen3 30B A3B Instruct 2507 is a July 2025 updated mixture-of-experts model from Alibaba with 30 billion total parameters but only around 3 billion active during inference. This MoE architecture gives it a remarkably small memory and compute footprint relative…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    93

    Q3_K_M · 16 GB

    188 tok/s

  11. 11

    Qwen3 32BAlibaba · 32.8B

    Qwen3 32B is the flagship dense model in Alibaba Cloud's Qwen 3 series, with 32 billion parameters. It is instruction-tuned for chat and delivers strong performance across reasoning, coding, mathematics, and multilingual tasks.

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    92

    Q4_K_M · 22.6 GB

    32 tok/s

  12. 12

    GPT OSS 20BOpenAI · 20.9B

    GPT-OSS 20B is one of OpenAI's first open-source model releases, marking a historic shift in the company's approach to open weights. At 21.5 billion parameters it delivers strong general-purpose chat and reasoning capabilities informed by the research behind the GPT…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    91

    IQ4_XS · 11.8 GB

    150 tok/s

  13. 13

    Qwen3 14BAlibaba · 14.8B

    Qwen3 14B is a 14-billion parameter instruction-tuned model from Alibaba Cloud's Qwen 3 series. It occupies a practical middle ground in the Qwen 3 lineup, offering stronger reasoning and generation quality than the 8B variant while remaining manageable on GPUs…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    88

    Q4_K_M · 11.9 GB

    46 tok/s

  14. 14

    Magistral Small 2506Mistral AI · 23.6B

    Magistral Small 2506 is Mistral AI's small, efficient reasoning model, built on the 24-billion-parameter Mistral Small 3.1 and further trained with supervised fine-tuning on Magistral Medium's reasoning traces plus reinforcement learning. It produces long chains of reasoning before answering and…

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    88

    Q3_K_L · 15.7 GB

    49 tok/s

  15. 15

    Qwen2.5 72B InstructAlibaba · 72.7B

    Qwen2.5 72B Instruct is the flagship model of the Qwen 2.5 series from Alibaba Cloud, with 72.7 billion parameters. It is instruction-tuned for conversational use and excels across reasoning, coding, mathematics, and multilingual tasks.

    • 8
    • 12
    • 16
    • 24
    • 32
    • 48

    86

    Q4_K_S · 46.6 GB

    15 tok/s

All 168 models tagged for reasoning
  • DeepSeek R1 Distill Qwen 1.5B8 GB+1.8 GB @Q4

    DeepSeek · 1.8B

    DeepSeek R1 Distill Qwen 1.5B is the smallest model in the R1 distillation family, packing chain-of-thought reasoning capabilities into just 1.5 billion parameters using the…

  • DeepSeek R196412.1 GB @Q4

    DeepSeek · 684.5B

    DeepSeek R1 is a groundbreaking reasoning model that uses reinforcement learning to develop chain-of-thought capabilities without relying on supervised fine-tuning.

  • Cosmos Reason2 2B8 GB+3.5 GB @Q4

    NVIDIA · 2.4B

    Cosmos Reason2 2B is NVIDIA's 2.4-billion-parameter open reasoning vision-language model for physical AI, built on the Qwen3-VL-2B-Instruct backbone and designed to let robots and embodied…

  • DeepSeek R1 0528 Qwen3 8B8 GB+7.6 GB @Q4

    DeepSeek · 8.2B

    With about 8 billion parameters, DeepSeek R1 0528 Qwen3 8B is a dated May 2025 release in DeepSeek's R1 line, built on a Qwen3-based architecture…

  • Ornith 1.0 35B AEON Ultimate Uncensored BF1624 GB+21.5 GB @Q4

    AEON-7 · 35.1B

    Ornith 1.0 35B AEON Ultimate Uncensored BF16 is a 35.1B-parameter open language model from AEON-7 in the Ornith family.

  • DeepSeek R1 Distill Qwen 32B24 GB+24.3 GB @Q4

    DeepSeek · 32.8B

    DeepSeek R1 Distill Qwen 32B takes the reasoning capabilities developed in the full 684.5B R1 model and distills them into the 32.8 billion parameter Qwen…

  • DeepSeek R1 Distill Qwen 7B8 GB+5.8 GB @Q4

    DeepSeek · 7.6B

    DeepSeek R1 Distill Qwen 7B compresses the reasoning techniques from DeepSeek's full R1 model into a compact 7.6 billion parameter dense model built on the…

  • Nemotron 3 Nano Omni 30B A3B Reasoning BF1624 GB+25.6 GB @Q4

    NVIDIA · 33.0B

    Nemotron 3 Nano Omni 30B A3B Reasoning BF16 is a 33.0B-parameter open language model from NVIDIA in the Nemotron family.

  • DeepSeek R1 Distill Qwen 14B12 GB+12.4 GB @Q4

    DeepSeek · 14.8B

    DeepSeek R1 Distill Qwen 14B sits in a sweet spot between the smaller 7B distill and the more demanding 32B version, offering strong reasoning performance…

  • Qwen3 8B 64k Context 2X Josiefied Uncensored8 GB+7.6 GB @Q4

    DavidAU · 8.2B

    Qwen3 8B 64k Context 2X Josiefied Uncensored is a 8.2B-parameter open language model from DavidAU in the Qwen 3 family.

  • TwIL LM38 GB+3.4 GB @Q4

    webAI-Official · 3.1B

    TwIL LM3 is a 3.1B-parameter open language model from webAI-Official.

  • DeepSeek R1 0528100412.2 GB @Q4

    DeepSeek · 684.5B

    DeepSeek R1 0528 is an updated release of the R1 reasoning model, incorporating improvements to training and inference that sharpen its performance on complex multi-step…

  • DeepSeek R1 Distill Llama 8B8 GB+7.3 GB @Q4

    DeepSeek · 8.0B

    DeepSeek R1 Distill Llama 8B brings R1's reinforcement-learned reasoning capabilities to the widely supported Llama 3.1 8B architecture.

  • Qwen3.8 4B Distill8 GB+3.4 GB @Q4

    empero-ai · 4.7B

    Qwen3.8 4B Distill is a 4.7B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • GLM 4.1V 9B Thinking8 GB+7.2 GB @Q4

    Z.ai · 10.3B

    GLM 4.1V 9B Thinking is a 10.3-billion-parameter vision-language reasoning model jointly released by Z.ai (Zhipu AI) and Tsinghua University's KEG lab, built on the GLM-4-9B-0414…

  • Cosmos Reason2 8B8 GB+8.4 GB @Q4

    NVIDIA · 8.8B

    Cosmos Reason2-8B is NVIDIA's 8.8-billion-parameter open reasoning vision-language model for physical AI, built on a Qwen3-VL-8B-Instruct backbone and tuned to reason step by step about…

  • Granite 4.2 8B8 GB+8.3 GB @Q4

    IBM · 8.8B

    Granite 4.2 8B is a 8.8B-parameter open language model from IBM in the Granite family.

  • Nemotron Cascade 2 30B A3B16 GB+19.3 GB @Q4

    NVIDIA · 31.6B

    Nemotron Cascade 2 30B A3B is a 31.6B-parameter open language model from NVIDIA in the Nemotron family.

  • Ministral 3 14B Reasoning 251212 GB+12.0 GB @Q4

    Mistral AI · 13.9B

    Ministral 3 14B Reasoning 2512 is a 13.9B-parameter open language model from Mistral AI in the Mistral family.

  • Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled16 GB+17.9 GB @Q4

    Jackrong · 27.8B

    The full-precision version of Jackrong's Qwen3.5 27B reasoning distillation from Claude 4.6 Opus.

  • VulnLLM R 7B8 GB+5.8 GB @Q4

    UCSB-SURFI · 7.6B

    VulnLLM R 7B is a security-focused model developed by UCSB-SURFI, built on the Qwen2.5-7B base and fine-tuned specifically for vulnerability analysis and security reasoning.

  • DeepSeek R1 Distill Llama 70B48 GB+48 GB @Q4

    DeepSeek · 70.6B

    DeepSeek R1 Distill Llama 70B is the largest model in the R1 distillation lineup, combining the reasoning capabilities developed in the full 684.5B R1 with…

  • Phi 4 Mini Reasoning8 GB+4.8 GB @Q4

    Microsoft · 3.8B

    Phi-4-mini-reasoning is Microsoft's compact 3.8-billion-parameter model in the Phi-4 family, fine-tuned specifically for multi-step, logic-intensive mathematical reasoning using synthetic math data distilled from a larger…

  • GLM 5.2 W4AFP8236.9 GB @Q4

    PhalaCloud · 391.9B

    GLM 5.2 W4AFP8 is a 391.9B-parameter open language model from PhalaCloud in the GLM 5 family.

  • Granite 4.2 3B8 GB+3.8 GB @Q4

    IBM · 3.7B

    Granite 4.2 3B is a 3.7B-parameter open language model from IBM in the Granite family.

  • AI21 Jamba Reasoning 3B8 GB+2.3 GB @Q4

    AI21 Labs · 3.2B

    AI21 Jamba Reasoning 3B is a 3.2B-parameter open language model from AI21 Labs in the Jamba family.

  • MN 12B Mag Mell R112 GB+11.0 GB @Q4

    inflatebot · 12.2B

    MN 12B Mag Mell R1 is a 12.2B-parameter open language model from inflatebot.

  • K2 Horizon 0.9B8 GB+2.2 GB @Q4

    IFM · 1.1B

    K2 Horizon 0.9B is a 1.1B-parameter open language model from IFM.

  • QwQ 32B24 GB+24.3 GB @Q4

    Alibaba · 32.8B

    QwQ 32B is a 32-billion parameter reasoning-focused model from Alibaba Cloud's Qwen family.

  • Granite 4.2 30B24 GB+22.2 GB @Q4

    IBM · 29.3B

    Granite 4.2 30B is a 29.3B-parameter open language model from IBM in the Granite family.

  • Mythos Nano8 GB+2.8 GB @Q4

    squ11z1 · 3.1B

    Mythos Nano is a 3.1B-parameter open language model from squ11z1.

  • OneReason 0.8B Pretrain Competition8 GB+1.7 GB @Q4

    OpenOneRec · 801M

    OneReason 0.8B Pretrain Competition is a 801M-parameter open language model from OpenOneRec.

  • Nemotron Cascade 8B8 GB+7.5 GB @Q4

    NVIDIA · 8B

    Nemotron Cascade 8B is a 8B-parameter open language model from NVIDIA in the Nemotron family.

  • Qwen3.6 35B A3B Claude 4.7 Opus Reasoning Distilled24 GB+22.0 GB @Q4

    lordx64 · 36.0B

    Qwen3.6 35B A3B Claude 4.7 Opus Reasoning Distilled is a 36.0B-parameter open language model from lordx64 in the Qwen 3.6 family.

  • Hermes 4.3 36B24 GB+24.1 GB @Q4

    Nous Research · 36.2B

    Hermes 4.3 36B is a 36.2B-parameter open language model from Nous Research in the Hermes family.

  • Phi 4 Reasoning12 GB+12.5 GB @Q4

    Microsoft · 14.7B

    Phi-4-reasoning is Microsoft's reasoning-focused fine-tune of Phi-4, a 14-billion-parameter dense decoder-only transformer, produced through supervised fine-tuning on chain-of-thought traces followed by reinforcement learning, concentrated on…

  • Phi 4 Reasoning Plus12 GB+12.5 GB @Q4

    Microsoft · 14.7B

    Phi 4 Reasoning Plus is a 14.7B-parameter open language model from Microsoft in the Phi 4 family.

  • Ouro 1.4B8 GB+2.5 GB @Q4

    ByteDance · 1.4B

    Ouro 1.4B is a 1.4B-parameter open language model from ByteDance.

  • Huginn 012512 GB+4.7 GB @Q4

    tomg-group-umd · 3.9B

    Huginn 0125 is a 3.9B-parameter open language model from tomg-group-umd.

  • Qwen3.8 27B EfficientThink Uncensored K3 Opus5 Grok4.6 GPT5.6Sol SFT SimPO DFlash224 GB+21.4 GB @Q4

    nerkyor · 27B

    Qwen3.8 27B EfficientThink Uncensored K3 Opus5 Grok4.6 GPT5.6Sol SFT SimPO DFlash2 is a 27B-parameter open language model from nerkyor in the Qwen 3.8 family.

  • Qwen3.8 27B Uncensored24 GB+22.0 GB @Q4

    orcarouter · 27.8B

    Qwen3.8 27B Uncensored is a 27.8B-parameter open language model from orcarouter in the Qwen 3.8 family.

  • Qwen3.8 9B8 GB+6.6 GB @Q4

    empero-ai · 9.7B

    Qwen3.8 9B is a 9.7B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • Foundation Sec 8B Reasoning8 GB+7.3 GB @Q4

    fdtn-ai · 8.0B

    Foundation Sec 8B Reasoning is a 8.0B-parameter open language model from fdtn-ai.

  • Kolibri 148 GB+47.9 GB @Q4

    Aleph-Alpha · 78.1B

    Kolibri 1 is a 78.1B-parameter open language model from Aleph-Alpha.

  • Ministral 3 3B Reasoning 25128 GB+4.2 GB @Q4

    Mistral AI · 4.3B

    Ministral 3 3B Reasoning 2512 is a 4.3B-parameter open language model from Mistral AI in the Mistral family.

  • Qwen3.8 4B8 GB+3.4 GB @Q4

    empero-ai · 4.7B

    Qwen3.8 4B is a 4.7B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • Qwen27b Abliterated Fable MTP24 GB+21.4 GB @Q4

    hotdogs · 27B

    Qwen27b Abliterated Fable MTP is a 27B-parameter open language model from hotdogs in the Qwen family.

  • QwQ 32B Preview24 GB+24.3 GB @Q4

    Alibaba · 32.8B

    QwQ 32B Preview is a 32.8B-parameter open language model from Alibaba in the QwQ family.

  • Limite 1B Violetto8 GB+2.1 GB @Q4

    paradigma-inc · 1.0B

    Limite 1B Violetto is a 1.0B-parameter open language model from paradigma-inc.

  • Ouro 2.6B Thinking16 GB+3.6 GB @Q4

    ByteDance · 2.7B

    Ouro 2.6B Thinking is a 2.7B-parameter open language model from ByteDance.

  • Qwen3.6 27B Uncensored HauhauCS Aggressive MTP24 GB+21.4 GB @Q4

    AIOpsInSpace · 27B

    Qwen3.6 27B Uncensored HauhauCS Aggressive MTP is a 27B-parameter open language model from AIOpsInSpace in the Qwen 3.6 family.

  • NeoHorse 1 9B8 GB+6.2 GB @Q4

    TokenRhythm · 9.0B

    NeoHorse 1 9B is a 9.0B-parameter open language model from TokenRhythm.

  • Qwopus3.6 27B Coder22.0 GB @Q4

    Jackrong · 27.8B

    Qwopus3.6 27B Coder is a 27.8B-parameter open language model from Jackrong.

  • Qwen3.8 27B AEON ULTIMATE UNCENSORED BF1622.0 GB @Q4

    AEON-7 · 27.8B

    Qwen3.8 27B AEON ULTIMATE UNCENSORED BF16 is a 27.8B-parameter open language model from AEON-7 in the Qwen 3.8 family.

  • NeoHorse 1 4B8 GB+3.2 GB @Q4

    TokenRhythm · 4.2B

    NeoHorse 1 4B is a 4.2B-parameter open language model from TokenRhythm.

  • Qwen3.8 2B8 GB+1.9 GB @Q4

    empero-ai · 2.3B

    Qwen3.8 2B is a 2.3B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • Jack 3.8 27B Coder 16GB VRAM21.4 GB @Q4

    JackAgentLead · 27B

    Jack 3.8 27B Coder 16GB VRAM is a 27B-parameter open language model from JackAgentLead.

  • Qwen3.8 9B Distill8 GB+6.6 GB @Q4

    empero-ai · 9.7B

    Qwen3.8 9B Distill is a 9.7B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • Ouro 1.4B Thinking8 GB+2.5 GB @Q4

    ByteDance · 1.4B

    Ouro 1.4B Thinking is a 1.4B-parameter open language model from ByteDance.

  • Qwen3.6 27B AEON Ultimate Uncensored BF1616 GB+17.6 GB @Q4

    AEON-7 · 27.4B

    Qwen3.6 27B AEON Ultimate Uncensored BF16 is a 27.4B-parameter open language model from AEON-7 in the Qwen 3.6 family.

  • Trinity Large Thinking270.8 GB @Q4

    Arcee AI · 398.6B

    Trinity Large Thinking is a 398.6B-parameter open language model from Arcee AI.

  • VibeThinker 3B8 GB+2.8 GB @Q4

    WeiboAI · 3.1B

    VibeThinker 3B is a 3.1B-parameter open language model from WeiboAI.

  • DeepSeek R1 Zero412.2 GB @Q4

    DeepSeek · 684.5B

    DeepSeek R1 Zero is a 684.5B-parameter open language model from DeepSeek in the DeepSeek R1 family.

  • Ouro 2.6B16 GB+3.6 GB @Q4

    ByteDance · 2.7B

    Ouro 2.6B is a 2.7B-parameter open language model from ByteDance.

  • Hermes 4 70B48 GB+48 GB @Q4

    Nous Research · 70.6B

    Hermes 4 70B is a 70.6B-parameter open language model from Nous Research in the Hermes family.

  • Qwen Marketing24 GB+8.0 GB @Q4

    marketeam · 8.2B

    Qwen Marketing is a 8.2B-parameter open language model from marketeam in the Qwen family.

  • Qwen2.5 Coder 3B Claude Opus 4.6 Distilled8 GB+2.8 GB @Q4

    ryzdfm · 3.1B

    Qwen2.5 Coder 3B Claude Opus 4.6 Distilled is a 3.1B-parameter open language model from ryzdfm in the Qwen 2.5 family.

  • Qwen3.8 35B A3B Distill24 GB+21.5 GB @Q4

    empero-ai · 35.1B

    Qwen3.8 35B A3B Distill is a 35.1B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • Swift 1.5 Qwen3.8 27B16 GB+17.9 GB @Q4

    ukisai · 27.8B

    Swift 1.5 Qwen3.8 27B is a 27.8B-parameter open language model from ukisai in the Qwen 3.8 family.

  • Huihui Qwen3.6 35B A3B Claude 4.7 Opus Abliterated24 GB+22.0 GB @Q4

    huihui-ai · 36.0B

    Huihui Qwen3.6 35B A3B Claude 4.7 Opus Abliterated is a 36.0B-parameter open language model from huihui-ai in the Qwen 3.6 family.

  • Qwythos 9B Claude Mythos 5 1M8 GB+6.5 GB @Q4

    empero-ai · 9.4B

    Qwythos 9B Claude Mythos 5 1M is a 9.4B-parameter open language model from empero-ai.

  • Tri 21B Think48 GB+17.0 GB @Q4

    trillionlabs · 20.7B

    Tri 21B Think is a 20.7B-parameter open language model from trillionlabs.

  • OFFELLIA Gemma 4 E4B 8B Claude 4.6 Opus Reasoning MTP8 GB+4.8 GB @Q4

    Brunobkr · 4B

    OFFELLIA Gemma 4 E4B 8B Claude 4.6 Opus Reasoning MTP is a 4B-parameter open language model from Brunobkr in the Gemma 4 family.

  • OrcaSAQ 2 27B16 GB+6.9 GB @Q4

    orcarouter · 6.8B

    OrcaSAQ 2 27B is a 6.8B-parameter open language model from orcarouter in the Orca family.

  • Qwen3.5 9B Claude 4.6 Opus Reasoning Distilled8 GB+6.6 GB @Q4

    Jackrong · 9.7B

    Qwen3.5 9B Claude 4.6 Opus Reasoning Distilled is a 9.7B-parameter open language model from Jackrong in the Qwen 3.5 family.

  • Whittle Next 27B A3B16 GB+17.5 GB @Q4

    logic65 · 27.6B

    Whittle Next 27B A3B is a 27.6B-parameter open language model from logic65.

  • Qwen3.8 27B Escha W216 GB+6.6 GB @Q4

    EschaLabs · 6.3B

    Qwen3.8 27B Escha W2 is a 6.3B-parameter open language model from EschaLabs in the Qwen 3.8 family.

  • TwIL LM3 Pro8 GB+3.8 GB @Q4

    webAI-Official · 3.7B

    TwIL LM3 Pro is a 3.7B-parameter open language model from webAI-Official.

  • Qwen3 4B Gemini 3.1 Pro Reasoning Distilled8 GB+4.2 GB @Q4

    khazarai · 4B

    Qwen3 4B Gemini 3.1 Pro Reasoning Distilled is a 4B-parameter open language model from khazarai in the Qwen 3 family.

  • Bonsai 2 27B CRACK Ternary JANG21.7 GB @Q4

    dealignai · 27.4B

    Bonsai 2 27B CRACK Ternary JANG is a 27.4B-parameter open language model from dealignai.

  • DeepSeek V4 Flash Vision Uncensored208.2 GB @Q4

    orcarouter · 304.6B

    DeepSeek V4 Flash Vision Uncensored is a 304.6B-parameter open language model from orcarouter in the DeepSeek V4 family.

  • Qwen3.6 35B A3B Claude 4.6 Opus Reasoning Distilled24 GB+22.0 GB @Q4

    hesamation · 36.0B

    Qwen3.6 35B A3B Claude 4.6 Opus Reasoning Distilled is a 36.0B-parameter open language model from hesamation in the Qwen 3.6 family.

  • Hermes 4 14B12 GB+11.8 GB @Q4

    Nous Research · 14.8B

    Hermes 4 14B is a 14.8B-parameter open language model from Nous Research in the Hermes family.

  • Qwen35B Agent R224 GB+21.3 GB @Q4

    hotdogs · 34.7B

    Qwen35B Agent R2 is a 34.7B-parameter open language model from hotdogs in the Qwen family.

  • INTELLECT 365.4 GB @Q4

    PrimeIntellect · 106.9B

    INTELLECT 3 is a 106.9B-parameter open language model from PrimeIntellect.

  • Cosmos Reason2 32B24 GB+23 GB @Q4

    NVIDIA · 33.4B

    Cosmos Reason 2 32B is NVIDIA's 32-billion-parameter reasoning vision-language model for physical AI and robotics, built on Qwen3-VL-32B-Instruct.

  • Qwen3.5 2B Claude 4.6 Opus Reasoning Distilled8 GB+1.9 GB @Q4

    Jackrong · 2.3B

    Qwen3.5 2B Claude 4.6 Opus Reasoning Distilled is a 2.3B-parameter open language model from Jackrong in the Qwen 3.5 family.

  • Qwen3.5 4B Safety Thinking8 GB+3.2 GB @Q4

    MerlinSafety · 4.2B

    Qwen3.5 4B Safety Thinking is a 4.2B-parameter open language model from MerlinSafety in the Qwen 3.5 family.

  • Qwen3.8 2B Distill8 GB+1.9 GB @Q4

    empero-ai · 2.3B

    Qwen3.8 2B Distill is a 2.3B-parameter open language model from empero-ai in the Qwen 3.8 family.

  • UltiMerge26.8 GB @Q4

    OliviaRossi · 34.7B

    UltiMerge is a 34.7B-parameter open language model from OliviaRossi.

  • Qwen3.5 4B Claude 4.6 Opus Reasoning Distilled8 GB+3.4 GB @Q4

    Jackrong · 4.7B

    Qwen3.5 4B Claude 4.6 Opus Reasoning Distilled is a 4.7B-parameter open language model from Jackrong in the Qwen 3.5 family.

  • Qwythos 9B v224 GB+9.1 GB @Q4

    empero-ai · 9.7B

    Qwythos 9B v2 is a 9.7B-parameter open language model from empero-ai.

  • Claude OSS8 GB+6.2 GB @Q4

    squ11z1 · 9.0B

    Claude OSS is a 9.0B-parameter open language model from squ11z1.

  • Nemotron Research Reasoning Qwen 1.5B8 GB+1.8 GB @Q4

    NVIDIA · 1.8B

    Nemotron Research Reasoning Qwen 1.5B is a 1.8B-parameter open language model from NVIDIA in the Qwen family.

  • Qwen3.5 35B A3B Claude 4.6 Opus Reasoning Distilled27.6 GB @Q4

    Jackrong · 36.0B

    Qwen3.5 35B A3B Claude 4.6 Opus Reasoning Distilled is a 36.0B-parameter open language model from Jackrong in the Qwen 3.5 family.

  • Maple Preview12 GB+13.2 GB @Q4

    deepgrove · 20.2B

    Maple Preview is a 20.2B-parameter open language model from deepgrove.

  • Kolibri 1 BF1648 GB+47.9 GB @Q4

    Aleph-Alpha · 78.1B

    Kolibri 1 stands out as Aleph Alpha's mixture-of-experts reasoning model, totaling 78 billion parameters with 3.46 billion active per token, targeting German and English.

  • Nemotron Labs Audex 2B8 GB+3.1 GB @Q4

    NVIDIA · 2B

    Nemotron Labs Audex 2B is a 2B-parameter open language model from NVIDIA in the Nemotron family.

  • Gemma 4 12B IT AEON Abliterated K4 BF168 GB+9.0 GB @Q4

    AEON-7 · 12.0B

    Gemma 4 12B IT AEON Abliterated K4 BF16 is a 12.0B-parameter open language model from AEON-7 in the Gemma 4 family.

  • Nemotron Content Safety Reasoning 4B8 GB+3.6 GB @Q4

    NVIDIA · 4.3B

    Nemotron Content Safety Reasoning 4B is a 4.3B-parameter open language model from NVIDIA in the Nemotron family.

  • Nemotron Labs Audex 30B A3B24 GB+23.5 GB @Q4

    NVIDIA · 30B

    Nemotron Labs Audex 30B A3B is a 30B-parameter open language model from NVIDIA in the Nemotron family.

  • Qwen3.8 27B Pi16 GB+17.9 GB @Q4

    bytkim · 27.8B

    Qwen3.8 27B Pi is a 27.8B-parameter open language model from bytkim in the Qwen 3.8 family.

  • Qwen3.5 9B The Defiant Fable Uncensored Heretic NEO IMATRIX MAX MTP8 GB+6.6 GB @Q4

    DavidAU · 9.7B

    Qwen3.5 9B The Defiant Fable Uncensored Heretic NEO IMATRIX MAX MTP is a 9.7B-parameter open language model from DavidAU in the Qwen 3.5 family.

  • ZGCM 1 7B24 GB+7.4 GB @Q4

    zgcagi · 7.4B

    ZGCM 1 7B is a 7.4B-parameter open language model from zgcagi.

  • Qwen3.8 Flash Next Uncensored124.9 GB @Q4

    orcarouter · 180.0B

    Qwen3.8 Flash Next Uncensored is a 180.0B-parameter open language model from orcarouter in the Qwen 3.8 family.

  • Carnice V1 9B Hermes Agent Stage2 Merged8 GB+6.2 GB @Q4

    kai-os · 9.0B

    Carnice V1 9B Hermes Agent Stage2 Merged is a 9.0B-parameter open language model from kai-os in the Hermes family.

  • Moderato V1 Pro80.2 GB @Q4

    nitrai-research · 113.3B

    Moderato V1 Pro is a 113.3B-parameter open language model from nitrai-research.

  • Qwen2.5 Coder 7B Bird Cot8 GB+5.8 GB @Q4

    jk200201 · 7.6B

    Qwen2.5 Coder 7B Bird Cot is a 7.6B-parameter open language model from jk200201 in the Qwen 2.5 family.

  • AREX 216 GB+17.6 GB @Q4

    BAAI · 27.4B

    AREX-2 is a 27-billion-parameter long-horizon agent model from the Beijing Academy of Artificial Intelligence (BAAI).

  • Cyber Ornith 1.5 9B OBLITERATED8 GB+6.2 GB @Q4

    DuoNeural · 9.0B

    Cyber Ornith 1.5 9B OBLITERATED is a 9.0B-parameter open language model from DuoNeural in the Ornith family.

  • Qwen3 42B A3B 2507 Thinking Abliterated Uncensored TOTAL RECALL v2 Medium MASTER CODER24 GB+26.8 GB @Q4

    DavidAU · 42.4B

    Qwen3 42B A3B 2507 Thinking Abliterated Uncensored TOTAL RECALL v2 Medium MASTER CODER is a 42.4B-parameter open language model from DavidAU in the Qwen 3…

  • Domyn Small v1.08 GB+8.0 GB @Q4

    domyn · 9.8B

    Domyn Small v1.0 is a 9.8B-parameter open language model from domyn.

  • Qwen3.8 9B Distill Uncensored Heretic8 GB+6.5 GB @Q4

    petruhonk · 9.4B

    Qwen3.8 9B Distill Uncensored Heretic is a 9.4B-parameter open language model from petruhonk in the Qwen 3.8 family.

  • Nex N2.5 Mini Uncensored24 GB+27.1 GB @Q4

    orcarouter · 35.1B

    Nex N2.5 Mini Uncensored is a 35.1B-parameter open language model from orcarouter.

  • Instinct Python Coder Gemma4 12B Qwen3.8 27B12 GB+10.8 GB @Q4

    projectj · 12B

    Instinct Python Coder Gemma4 12B Qwen3.8 27B is a 12B-parameter open language model from projectj in the Qwen 3.8 family.

  • Agnes 3.0 Flash24 GB+24.2 GB @Q4

    Agnes-AI · 33.1B

    Agnes-3.0-Flash Preview is Agnes AI's 33-billion-parameter open-weight multimodal model offering competitive reasoning, coding, and instruction-following at lower hardware cost than flagship-class models; it understands text…

  • Turkish Gemma 9B T18 GB+7.7 GB @Q4

    ytu-ce-cosmos · 9.2B

    Turkish Gemma 9B T1 is a 9.2B-parameter open language model from ytu-ce-cosmos in the Gemma family.

  • Parable Qwen3 4B Claude Fable 58 GB+4.2 GB @Q4

    AnkitAI · 4.0B

    Parable Qwen3 4B Claude Fable 5 is a 4.0B-parameter open language model from AnkitAI in the Qwen 3 family.

  • Qwopus3.5 4B Coder12 GB+5.3 GB @Q4

    Jackrong · 4.7B

    Qwopus3.5 4B Coder is a 4.7B-parameter open language model from Jackrong.

  • ExtGemma4 40 5B30.1 GB @Q4

    TOTORONG · 39.5B

    ExtGemma4 40 5B is a 39.5B-parameter open language model from TOTORONG in the Gemma 4 family.

  • SuperDeepseek V4 Flash Abliterated MQ 2xDGX184.2 GB @Q4

    Jiunsong · 306.3B

    SuperDeepseek V4 Flash Abliterated MQ 2xDGX is a 306.3B-parameter open language model from Jiunsong in the DeepSeek V4 family.

  • Qwen3.5 4B Claude Opus 4.6 Distilled Heretic12 GB+5.2 GB @Q4

    ghost-actual · 4.5B

    Qwen3.5 4B Claude Opus 4.6 Distilled Heretic is a 4.5B-parameter open language model from ghost-actual in the Qwen 3.5 family.

  • Grug V1.1 Qwen 3.8 27B21.7 GB @Q4

    ProCreations · 27.4B

    Grug V1.1 Qwen 3.8 27B is a 27.4B-parameter open language model from ProCreations in the Qwen 3.8 family.

  • Grug 12B32 GB+10.8 GB @Q4

    kai-os · 12.0B

    Grug 12B is a 12.0B-parameter open language model from kai-os.

  • Raptor 0.6 4B JANG 6M12 GB+4.8 GB @Q4

    OsaurusAI · 4.1B

    Raptor 0.6 4B JANG 6M is a 4.1B-parameter open language model from OsaurusAI.

  • Soren 1 Small8 GB+3.0 GB @Q4

    syntropy-ai · 1.9B

    Soren 1 Small is a 1.9B-parameter open language model from syntropy-ai.

  • Qwen3.6 27B AEON Ultimate Uncensored16 GB+17.6 GB @Q4

    AEON-7 · 27.4B

    Qwen3.6 27B AEON Ultimate Uncensored is a 27.4B-parameter open language model from AEON-7 in the Qwen 3.6 family.

  • Grug 35B A3B26.8 GB @Q4

    kai-os · 34.7B

    Grug 35B A3B is a 34.7B-parameter open language model from kai-os.

  • Parable Granite 4.1 8B Claude Fable 58 GB+8.0 GB @Q4

    AnkitAI · 8.4B

    Parable Granite 4.1 8B Claude Fable 5 is a 8.4B-parameter open language model from AnkitAI in the Granite family.

  • Raptor V0.5 8B A1B JANG 6M16 GB+6.4 GB @Q4

    OsaurusAI · 6.2B

    Raptor V0.5 8B A1B JANG 6M is a 6.2B-parameter open language model from OsaurusAI.

  • Darwin 27B RSI21.4 GB @Q4

    FINAL-Bench · 26.9B

    Darwin 27B RSI is a 26.9B-parameter open language model from FINAL-Bench.

  • Grug 27B21.7 GB @Q4

    ProCreations · 27.4B

    Grug 27B is a 27.4B-parameter open language model from ProCreations.

  • ViAble 1.7 9B24 GB+8.6 GB @Q4

    bogairff55 · 9.0B

    ViAble 1.7 9B is a 9.0B-parameter open language model from bogairff55.

  • Minicpm5 2B Distilled Reasoning8 GB+2.5 GB @Q4

    jigs97022 · 2.5B

    Minicpm5 2B Distilled Reasoning is a 2.5B-parameter open language model from jigs97022 in the MiniCPM family.

  • Phi 4 Mini Flash Reasoning8 GB+5.3 GB @Q4

    Microsoft · 3.9B

    Phi 4 Mini Flash Reasoning is a 3.9B-parameter open language model from Microsoft in the Phi 4 family.

  • Qwen35b Agent R2O324 GB+21.3 GB @Q4

    hotdogs · 34.7B

    Qwen35b Agent R2O3 is a 34.7B-parameter open language model from hotdogs in the Qwen family.

  • DeepSeek R1 Distill Qwen 32B Abliterated24 GB+24.3 GB @Q4

    huihui-ai · 32.8B

    DeepSeek R1 Distill Qwen 32B Abliterated is a 32.8B-parameter open language model from huihui-ai in the DeepSeek R1 family.

  • Qwen 3.5 Sql 27B Distill 2B21.4 GB @Q4

    AlioLeuchtmann · 27B

    Qwen 3.5 Sql 27B Distill 2B is a 27B-parameter open language model from AlioLeuchtmann in the Qwen 3.5 family.

  • Qwen3.5 4B EmperoAI Qwen3.8 Distill Heretic Abliterated12 GB+5.2 GB @Q4

    insraq · 4.5B

    Qwen3.5 4B EmperoAI Qwen3.8 Distill Heretic Abliterated is a 4.5B-parameter open language model from insraq in the Qwen 3.5 family.

  • Gemma 4 12B Agentic Fable5 Composer2.5 v2 3.5x Tau28 GB+9.0 GB @Q4

    yuxinlu1 · 12.0B

    Gemma 4 12B Agentic Fable5 Composer2.5 v2 3.5x Tau2 is a 12.0B-parameter open language model from yuxinlu1 in the Gemma 4 family.

  • OpenReasoning Nemotron 32B24 GB+24.3 GB @Q4

    NVIDIA · 32.8B

    OpenReasoning Nemotron 32B is a 32.8B-parameter open language model from NVIDIA in the Nemotron family.

  • DeepSeek R1 Distill Qwen 14B Abliterated v212 GB+12.4 GB @Q4

    huihui-ai · 14.8B

    DeepSeek R1 Distill Qwen 14B Abliterated v2 is a 14.8B-parameter open language model from huihui-ai in the DeepSeek R1 family.

  • Darwin 60B DUO44.1 GB @Q4

    FINAL-Bench · 60B

    Darwin 60B DUO is a 60B-parameter open language model from FINAL-Bench.

  • Nemotron H 8B Reasoning 128K24 GB+7.9 GB @Q4

    NVIDIA · 8.1B

    Nemotron H 8B Reasoning 128K is a 8.1B-parameter open language model from NVIDIA in the Nemotron family.

  • Hermes 4 405B252.3 GB @Q4

    Nous Research · 405.9B

    Hermes 4 405B is a 405.9B-parameter open language model from Nous Research in the Hermes family.

  • Qwen3.5 9B Gemini 3.1 Pro Reasoning Distill8 GB+6.6 GB @Q4

    Jackrong · 9.7B

    Qwen3.5 9B Gemini 3.1 Pro Reasoning Distill is a 9.7B-parameter open language model from Jackrong in the Qwen 3.5 family.

  • Kai 30B Instruct24 GB+24.3 GB @Q4

    NoesisLab · 32.8B

    Kai 30B Instruct is a 32.8B-parameter open language model from NoesisLab.

  • Supra 50M Reasoning8 GB+0.3 GB @Q4

    SupraLabs · 52M

    Supra 50M Reasoning is a 52M-parameter open language model from SupraLabs.

  • Darwin 4B Genesis16 GB+7.5 GB @Q4

    FINAL-Bench · 7.5B

    Darwin 4B Genesis is a 7.5B-parameter open language model from FINAL-Bench.

  • Beam 1.0 501B A23B304.0 GB @Q4

    reflection · 501.3B

    Beam 1.0 is an open model from Reflection, a sparse mixture-of-experts language model with about 501 billion total parameters and 23 billion active per token.

  • Darwin 36B Opus24 GB+21.3 GB @Q4

    FINAL-Bench · 34.7B

    Darwin 36B Opus is a 34.7B-parameter open language model from FINAL-Bench.

  • Qwopus3.5 9B v38 GB+6.6 GB @Q4

    Jackrong · 9.7B

    Qwopus3.5-9B-v3 is a community fine-tune by Jackrong, built on Qwen3.5-9B, aimed at steadier and more efficient reasoning, especially for programming.

  • Qwopus3.5 9B V3.58 GB+6.6 GB @Q4

    Jackrong · 9.7B

    Qwopus3.5 9B V3.5 is a 9.7B-parameter open language model from Jackrong.

  • CyberPal2.0 20B12 GB+13.1 GB @Q4

    cyber-pal-security · 20.9B

    CyberPal2.0 20B is a 20.9B-parameter open language model from cyber-pal-security.

  • Nemotron H 47B Reasoning 128K35.1 GB @Q4

    NVIDIA · 46.8B

    Nemotron H 47B Reasoning 128K is a 46.8B-parameter open language model from NVIDIA in the Nemotron family.

  • Aryabhata 2.012 GB+13.1 GB @Q4

    PhysicsWallahAI · 20.9B

    Aryabhata 2.0 is a 20.9B-parameter open language model from PhysicsWallahAI.

  • Datarus R1 14B Preview12 GB+12.4 GB @Q4

    DatarusAI · 14.8B

    Datarus R1 14B Preview is a 14.8B-parameter open language model from DatarusAI.

  • Qwen3 Code Reasoning 4B8 GB+4.2 GB @Q4

    GetSoloTech · 4B

    Qwen3 Code Reasoning 4B is a 4B-parameter open language model from GetSoloTech in the Qwen 3 family.

  • Scout 4B8 GB+3.6 GB @Q4

    vanta-research · 4.3B

    Scout 4B is a 4.3B-parameter open language model from vanta-research.

  • Turkish Gemma 4B T1 Scout8 GB+3.6 GB @Q4

    ytu-ce-cosmos · 4.3B

    Turkish Gemma 4B T1 Scout is a 4.3B-parameter open language model from ytu-ce-cosmos in the Gemma 4 family.

  • LFM2.5 8B A1B Opus Distil8 GB+5.6 GB @Q4

    reaperdoesntknow · 8.5B

    LFM2.5 8B A1B Opus Distil is a 8.5B-parameter open language model from reaperdoesntknow in the LFM2.5 family.

  • MAI DS R1404.1 GB @Q4

    Microsoft · 671.0B

    MAI DS R1 is a 671.0B-parameter open language model from Microsoft.

  • MIST Mini 8B Thinking8 GB+7.3 GB @Q4

    olaverse · 8.0B

    MIST Mini 8B Thinking is a 8.0B-parameter open language model from olaverse.

  • Pantheon Reasoning 27B16 GB+17.9 GB @Q4

    Gryphe · 27.8B

    Pantheon Reasoning 27B is a 27.8B-parameter open language model from Gryphe.

  • Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Heretic v216 GB+17.6 GB @Q4

    llmfan46 · 27.4B

    Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Heretic v2 is a 27.4B-parameter open language model from llmfan46 in the Qwen 3.5 family.

  • OpenCodeReasoning Nemotron 1.1 32B24 GB+24.3 GB @Q4

    NVIDIA · 32.8B

    OpenCodeReasoning Nemotron 1.1 32B is a 32.8B-parameter open language model from NVIDIA in the Nemotron family.

  • MobileLLM R1.5 950M8 GB+2.0 GB @Q4

    Meta · 950M

    MobileLLM R1.5 950M is a 950M-parameter open language model from Meta.

  • WorldSim Opus 3.6 35B A3B27.1 GB @Q4

    Gryphe · 35.1B

    WorldSim Opus 3.6 35B A3B is a 35.1B-parameter open language model from Gryphe.

What hardware do reasoning models need?

Reasoning models generate long chains of thought, so context length and generation speed matter as much as raw VRAM. A distilled 7B–8B reasoner (e.g. DeepSeek-R1-Distill-Qwen-7B) fits an 8 GB GPU; a 32B reasoner like QwQ-32B wants a 24 GB card; and the largest MoE reasoners run best on multi-GPU or high-memory Apple Silicon. Because reasoning models emit many tokens, faster memory bandwidth noticeably improves the experience. Open any model to see its VRAM-by-quantization table and estimated tokens/sec on your hardware.

Frequently Asked Questions

What is the best local reasoning LLM in 2026?

DeepSeek-R1 distills (the 7B/8B variants for consumer GPUs) and QwQ-32B are among the strongest reasoning models you can run locally. For maximum capability, larger MoE reasoners exist but need serious hardware. Open any model below to confirm it fits your GPU or Mac.

Can I run DeepSeek R1 locally?

The full DeepSeek R1 is very large, but its distilled versions (DeepSeek-R1-Distill-Qwen-7B, -Llama-8B, -Qwen-32B) are designed to run on consumer hardware — the 7B/8B distills fit a single 8–12 GB GPU at Q4_K_M.

Do reasoning models need more VRAM than regular models?

Not for the weights — VRAM for weights depends on size and quantization like any model. But reasoning models produce long outputs, so allow extra VRAM for the KV cache at long context, and prefer hardware with high memory bandwidth for faster generation.