Local AI Advisor

Model catalog

Every open-weight model the advisor can evaluate, with the specs that decide whether it will run comfortably on your machine.

Showing 48 of 48 models

  • DeepSeek-V4.1-Flash (748B-A16B)

    DeepSeek · DeepSeek-V4

    MoEGeneral Vision ThinkingOpen source
    Parameters
    748B / 16B active
    Context window
    1M tokens
    Weights at Q4_K_M
    427 GB
    Tool calling
    Reliable
    License
    MIT
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-17low confidence
    Can my computer run it? What hardware do I need? Compare

  • Kolibri-1 (78B-A3.5B)

    Aleph Alpha · Kolibri

    MoEGeneral ThinkingOpen source
    Parameters
    78B / 3.5B active
    Context window
    256K tokens
    Weights at Q4_K_M
    45 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Advertised up to 1M context; vendor recommends ≤256K for serving efficiency.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • LFM2.5 24B-A2B

    Liquid AI · LFM2.5

    MoEReasoning Thinking
    Parameters
    24B / 2.0B active
    Context window
    128K tokens
    Weights at Q4_K_M
    14 GB
    Tool calling
    Good
    License
    LFM Open License v1.0
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Always thinks before answering; supports function calling, post-trained with agentic RL in OpenClaw/Hermes/Pi/OpenCode harnesses. 64 experts, top-4 routing. Capability tiers are conservative placeholders pending independent benchmarks.

    huggingface.co(opens in a new tab)verified 2026-10-18high confidence
    Can my computer run it? What hardware do I need? Compare

  • Muse Spark Open 816B-A42B

    Meta · Muse Spark

    MoEReasoning Vision Thinking
    Parameters
    816B / 42B active
    Context window
    1M tokens
    Weights at Q4_K_M
    470 GB
    Tool calling
    Reliable
    License
    Muse Model License
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Natively multimodal (text/image/video/audio in, text out) via a 1.8B perception encoder; 256 experts, 8 active per token; reasoning effort low→xhigh (default xhigh). Official NVFP4 + BF16 checkpoints; community GGUF Q2_K (304 GB) / Q4_K_S (470 GB) / Q8_0 (872 GB). Meta's card asks ≥700 GB GPU memory for NVFP4, 2 TB for BF16 (fits 2x 512 GB M5 Ultra Mac Studios via MLX). License requires 'Built with Muse Spark Open' attribution; hosted businesses over $20M revenue need a separate agreement. Capability tiers are conservative placeholders pending independent benchmarks — flagged for weekly audit.

    huggingface.co(opens in a new tab)verified 2026-10-16high confidence
    Can my computer run it? What hardware do I need? Compare

  • Nemotron 3.5 Bolt 74B-A9B

    NVIDIA · Nemotron 3.5

    MoEVision Vision Thinking
    Parameters
    74B / 9.0B active
    Context window
    256K tokens
    Weights at Q4_K_M
    40 GB
    Tool calling
    Good
    License
    OpenMDW-1.1
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Distilled from Super VL with Star Elastic (40 MoE layers, 240–496 experts/layer, top-9); runs through llama.cpp/vLLM/Ollama/LM Studio. Headline release is a native 2-bit (IQ2) GGUF at ~24 GB; the schema has no 2-bit quant id, so the catalog lists the verified community MLX 4-bit build (Duckets, 40.2 GiB, runs on Apple Silicon/LM Studio) plus official BF16. Reasoning toggled via chat template; ships with a DFlash speculative drafter. Capability tiers are conservative placeholders pending independent benchmarks — flagged for weekly audit.

    huggingface.co(opens in a new tab)verified 2026-10-16high confidence
    Can my computer run it? What hardware do I need? Compare

  • Nemotron 3.5 Super VL 120B-A12B

    NVIDIA · Nemotron 3.5

    MoEVision Vision Thinking
    Parameters
    120B / 12B active
    Context window
    1M tokens
    Weights at MXFP4
    60 GB
    Tool calling
    Good
    License
    OpenMDW-1.1
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Nemotron-H hybrid: interleaved Mamba-2 + MoE layers (512 routed experts, top-22) with C-RADIO vision encoder; native MTP speculative decoding; reasoning toggled via chat template. Official NVFP4 + BF16 checkpoints (mxfp4 id covers NVIDIA's NVFP4 build). NVIDIA's benchmark wins are vendor-reported on specialized agentic tasks. Capability tiers are conservative placeholders pending independent benchmarks — flagged for weekly audit.

    huggingface.co(opens in a new tab)verified 2026-10-16high confidence
    Can my computer run it? What hardware do I need? Compare

  • Reflection Beam (501B-A23B)

    Reflection AI · Beam

    MoEReasoning ThinkingOpen source
    Parameters
    501B / 23B active
    Context window
    1M tokens
    Weights at MXFP4
    263 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Text-only MoE (256 experts, 8 active/token) aimed at agentic tool use and coding with token-efficient reasoning; MCP tool integration and browsing. Official FP16 + FP8 + NVFP4 checkpoints; day-0 support in Ollama, LM Studio and major inference clouds. Apache 2.0. Capability tiers are conservative placeholders pending independent benchmarks — flagged for weekly audit.

    huggingface.co(opens in a new tab)verified 2026-10-18high confidence
    Can my computer run it? What hardware do I need? Compare

  • Step 5 Preview 600B-A27B

    Stepfun · Step 5

    MoEReasoning Vision Thinking
    Parameters
    600B / 27B active
    Context window
    1M tokens
    Weights at F16 / BF16
    1129 GB
    Tool calling
    Good
    License
    StepFun Model License 1.0
    Released
    2026-10

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Sparse MoE (352 experts, top-8) with a sliding+full attention mix (every 4th layer full attention) and a perception encoder; reasoning effort modes low→xhigh; official BF16 + FP8 checkpoints (FP8: ~600 GB min GPU memory per the model card). No. 2 on OpenRouter's weekly usage chart at release (~18.8T tokens in 7 days). License is Apache-2.0-style with attribution and commercial conditions. Capability tiers are conservative placeholders pending independent benchmarks — flagged for weekly audit.

    huggingface.co(opens in a new tab)verified 2026-10-18high confidence
    Can my computer run it? What hardware do I need? Compare

  • MiniCPM5 2B

    OpenBMB · MiniCPM5

    DenseGeneralOpen source
    Parameters
    2.5B
    Context window
    128K tokens
    Weights at Q4_K_M
    1.5 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-09

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Second model in the MiniCPM5 series (after MiniCPM5-1B); vendor claims 2B-class open-source SOTA, competitive with 4B-class models.

    huggingface.co(opens in a new tab)verified 2026-10-11high confidence
    Can my computer run it? What hardware do I need? Compare

  • GLM-5.3 (753B-A40B)

    Z.ai · GLM

    MoEGeneral Thinking
    Parameters
    753B / 40B active
    Context window
    1M tokens
    Weights at Q4_K_M
    435 GB
    Tool calling
    Reliable
    License
    GLM-5.3
    Released
    2026-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • GLM-5.3-Flash (320B-A18B)

    Z.ai · GLM

    MoEGeneral Vision ThinkingOpen source
    Parameters
    320B / 18B active
    Context window
    1M tokens
    Weights at Q4_K_M
    186 GB
    Tool calling
    Reliable
    License
    MIT
    Released
    2026-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Muse Glimmer (30B)

    Meta · Muse Glimmer

    DenseGeneral Vision ThinkingOpen source
    Parameters
    30B
    Context window
    128K tokens
    Weights at Q4_K_M
    16 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    30B = 27.8B dense decoder + ~1.8B ViT-G/14 perception encoder. Official GGUF repo also ships a 1.4 GB mmproj (required for image input) and an optional 1.6 GB DFlash speculative-decoding drafter.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.8 27B

    Alibaba Qwen · Qwen3.8

    DenseGeneral Vision ThinkingOpen source
    Parameters
    28B
    Context window
    256K tokens
    Weights at Q4_K_M
    15 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Hybrid architecture: 3 of 4 layers use linear attention, so the KV cache grows ~4× slower than a classic 27B model.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Gemma 4 12B

    Google · Gemma 4

    DenseGeneral VisionOpen source
    Parameters
    12B
    Context window
    256K tokens
    Weights at Q4_K_M
    6.6 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-05

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • DeepSeek-V4-Flash (284B-A13B)

    DeepSeek · DeepSeek-V4

    MoEGeneral ThinkingOpen source
    Parameters
    284B / 13B active
    Context window
    1M tokens
    Weights at Q4_K_M
    162 GB
    Tool calling
    Reliable
    License
    MIT
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Frontier-class open model that only fits on very large unified-memory machines. KV-cache size is a rough estimate (compressed sparse attention).

    huggingface.co(opens in a new tab)verified 2026-10-08low confidence
    Can my computer run it? What hardware do I need? Compare

  • Gemma 4 26B-A4B

    Google · Gemma 4

    MoEGeneral VisionOpen source
    Parameters
    26B / 3.8B active
    Context window
    256K tokens
    Weights at Q4_K_M
    16 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Gemma 4 31B

    Google · Gemma 4

    DenseGeneral VisionOpen source
    Parameters
    31B
    Context window
    256K tokens
    Weights at Q4_K_M
    17 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Gemma 4 E4B

    Google · Gemma 4

    DenseGeneral VisionOpen source
    Parameters
    8.0B
    Context window
    128K tokens
    Weights at Q4_K_M
    4.6 GB
    Tool calling
    Basic
    License
    Apache-2.0
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    ~4.5B effective parameters per token; 8B including per-layer embeddings.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.6 27B

    Alibaba Qwen · Qwen3.6

    DenseGeneral Vision ThinkingOpen source
    Parameters
    28B
    Context window
    256K tokens
    Weights at Q4_K_M
    16 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.6 35B-A3B

    Alibaba Qwen · Qwen3.6

    MoEGeneral Vision ThinkingOpen source
    Parameters
    36B / 3.0B active
    Context window
    256K tokens
    Weights at Q4_K_M
    21 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Only ~3B parameters are active per token, so it generates like a small model while needing memory like a 35B one.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Mistral Small 4 (119B-A6.5B)

    Mistral AI · Mistral Small

    MoEGeneral VisionOpen source
    Parameters
    119B / 6.5B active
    Context window
    256K tokens
    Weights at Q4_K_M
    69 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-03

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Multi-head latent attention keeps the KV cache very small.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.5 122B-A10B

    Alibaba Qwen · Qwen3.5

    MoEGeneral Vision ThinkingOpen source
    Parameters
    125B / 10B active
    Context window
    256K tokens
    Weights at Q4_K_M
    71 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-02

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.5 27B

    Alibaba Qwen · Qwen3.5

    DenseGeneral Vision ThinkingOpen source
    Parameters
    28B
    Context window
    256K tokens
    Weights at Q4_K_M
    16 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-02

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.5 35B-A3B

    Alibaba Qwen · Qwen3.5

    MoEGeneral Vision ThinkingOpen source
    Parameters
    36B / 3.0B active
    Context window
    256K tokens
    Weights at Q4_K_M
    21 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2026-02

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.5 4B

    Alibaba Qwen · Qwen3.5

    DenseGeneral Vision ThinkingOpen source
    Parameters
    4.7B
    Context window
    256K tokens
    Weights at Q4_K_M
    2.6 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-02

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3.5 9B

    Alibaba Qwen · Qwen3.5

    DenseGeneral Vision ThinkingOpen source
    Parameters
    9.7B
    Context window
    256K tokens
    Weights at Q4_K_M
    5.3 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2026-02

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • GLM-4.7-Flash (30B-A3B)

    Z.ai · GLM

    MoECoder ThinkingOpen source
    Parameters
    31B / 3.0B active
    Context window
    198K tokens
    Weights at Q4_K_M
    17 GB
    Tool calling
    Reliable
    License
    MIT
    Released
    2026-01

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Devstral Small 2 (24B)

    Mistral AI · Devstral

    DenseCoder VisionOpen source
    Parameters
    24B
    Context window
    256K tokens
    Weights at Q4_K_M
    13 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2025-12

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Built for agentic software-engineering tools.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • GLM-4.6V-Flash (10B)

    Z.ai · GLM

    DenseVision VisionOpen source
    Parameters
    10B
    Context window
    128K tokens
    Weights at Q4_K_M
    5.8 GB
    Tool calling
    Good
    License
    MIT
    Released
    2025-12

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3-Next 80B-A3B Instruct

    Alibaba Qwen · Qwen3-Next

    MoEGeneralOpen source
    Parameters
    81B / 3.0B active
    Context window
    256K tokens
    Weights at Q4_K_M
    45 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-09

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • gpt-oss-120b

    OpenAI · gpt-oss

    MoEReasoning ThinkingOpen source
    Parameters
    117B / 5.1B active
    Context window
    128K tokens
    Weights at MXFP4
    59 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • gpt-oss-20b

    OpenAI · gpt-oss

    MoEReasoning ThinkingOpen source
    Parameters
    21B / 3.6B active
    Context window
    128K tokens
    Weights at MXFP4
    11 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-08

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Released natively in MXFP4 — further quantization barely shrinks it.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • GLM-4.5-Air (106B-A12B)

    Z.ai · GLM

    MoEGeneral ThinkingOpen source
    Parameters
    111B / 12B active
    Context window
    128K tokens
    Weights at Q4_K_M
    68 GB
    Tool calling
    Reliable
    License
    MIT
    Released
    2025-07

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3 30B-A3B Instruct 2507

    Alibaba Qwen · Qwen3

    MoEGeneralOpen source
    Parameters
    31B / 3.3B active
    Context window
    256K tokens
    Weights at Q4_K_M
    17 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-07

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3-Coder 30B-A3B Instruct

    Alibaba Qwen · Qwen3-Coder

    MoECoderOpen source
    Parameters
    31B / 3.3B active
    Context window
    256K tokens
    Weights at Q4_K_M
    17 GB
    Tool calling
    Reliable
    License
    Apache-2.0
    Released
    2025-07

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Mistral Small 3.2 (24B)

    Mistral AI · Mistral Small

    DenseGeneral VisionOpen source
    Parameters
    24B
    Context window
    128K tokens
    Weights at Q4_K_M
    13 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-06

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • DeepSeek-R1-0528-Qwen3-8B

    DeepSeek · DeepSeek-R1

    DenseReasoning ThinkingOpen source
    Parameters
    8.2B
    Context window
    128K tokens
    Weights at Q4_K_M
    4.7 GB
    Tool calling
    Basic
    License
    MIT
    Released
    2025-05

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Llama 4 Scout (109B-A17B)

    Meta · Llama 4

    MoEGeneral VisionOpen weights (restricted)
    Parameters
    109B / 17B active
    Context window
    256K tokens
    Weights at Q4_K_M
    61 GB
    Tool calling
    Good
    License
    Llama 4 Community License
    Released
    2025-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    Advertised up to 10M context; we cap practical planning at 256K.

    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3 14B

    Alibaba Qwen · Qwen3

    DenseGeneral ThinkingOpen source
    Parameters
    15B
    Context window
    32K tokens
    Weights at Q4_K_M
    8.4 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3 32B

    Alibaba Qwen · Qwen3

    DenseGeneral ThinkingOpen source
    Parameters
    33B
    Context window
    32K tokens
    Weights at Q4_K_M
    18 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing

    32K native context; 128K with YaRN scaling.

    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen3 8B

    Alibaba Qwen · Qwen3

    DenseGeneral ThinkingOpen source
    Parameters
    8.2B
    Context window
    32K tokens
    Weights at Q4_K_M
    4.7 GB
    Tool calling
    Good
    License
    Apache-2.0
    Released
    2025-04

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Gemma 3 27B

    Google · Gemma 3

    DenseGeneral VisionOpen weights (restricted)
    Parameters
    27B
    Context window
    128K tokens
    Weights at Q4_K_M
    15 GB
    Tool calling
    Basic
    License
    Gemma Terms of Use
    Released
    2025-03

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Llama 3.3 70B Instruct

    Meta · Llama 3

    DenseGeneralOpen weights (restricted)
    Parameters
    71B
    Context window
    128K tokens
    Weights at Q4_K_M
    40 GB
    Tool calling
    Good
    License
    Llama 3.3 Community License
    Released
    2024-12

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Phi-4 (14B)

    Microsoft · Phi

    DenseReasoningOpen source
    Parameters
    15B
    Context window
    16K tokens
    Weights at Q4_K_M
    8.3 GB
    Tool calling
    Basic
    License
    MIT
    Released
    2024-12

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08high confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen2.5-Coder 32B Instruct

    Alibaba Qwen · Qwen2.5-Coder

    DenseCoderOpen source
    Parameters
    33B
    Context window
    32K tokens
    Weights at Q4_K_M
    19 GB
    Tool calling
    Basic
    License
    Apache-2.0
    Released
    2024-11

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Llama 3.2 3B Instruct

    Meta · Llama 3

    DenseChatOpen weights (restricted)
    Parameters
    3.2B
    Context window
    128K tokens
    Weights at Q4_K_M
    1.9 GB
    Tool calling
    Basic
    License
    Llama 3.2 Community License
    Released
    2024-09

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Qwen2.5-Coder 7B Instruct

    Alibaba Qwen · Qwen2.5-Coder

    DenseCoderOpen source
    Parameters
    7.6B
    Context window
    32K tokens
    Weights at Q4_K_M
    4.4 GB
    Tool calling
    Basic
    License
    Apache-2.0
    Released
    2024-09

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare

  • Llama 3.1 8B Instruct

    Meta · Llama 3

    DenseChatOpen weights (restricted)
    Parameters
    8.0B
    Context window
    128K tokens
    Weights at Q4_K_M
    4.6 GB
    Tool calling
    Basic
    License
    Llama 3.1 Community License
    Released
    2024-07

    Capability tiers · editorial assessment, not benchmark scores

    • General
    • Coding
    • Reasoning
    • Agentic
    • Long context
    • Writing
    huggingface.co(opens in a new tab)verified 2026-10-08medium confidence
    Can my computer run it? What hardware do I need? Compare