02 Target GPUs18 VERIFIED ENGINES

Best Local AI Models for 6GB VRAM GPUs (GTX 1660, RTX 3050 & Entry Laptops)

Verified ultra-compact open-weight models and GGUF Q4 builds that run locally inside a 6GB VRAM ceiling or on CPU without out-of-memory crashes.

TARGET VRAM6 GB GDDR6

REPRESENTATIVE GPUSGTX 1660 · RTX 3050

TOP FITTING MODELSKokoro-82M & Wan 2.2 Q4

SIBLING HUB URLS:/hubs/target-gpus /vram/6gb /vram/8gb /vram/12gb /vram/16gb /vram/24gb /vram/32gb /vram/48gb /vram/80gb

Verified Engines for /vram/6gb

18 RESULTSClick any model card to open its dedicated /ai/[slug] page.

●LOCAL • 86/100

Llama 3.1 8B Instruct

Meta

Ultra-compact 3B open-weight model engineered for on-device summarization, instruction following, and local agent tasks on phones and 4GB laptops.

BASE ARCHITECTURE / Llama-3.2-3B-Instruct

SRuns well (95/100) • Q2_K (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

The best lightweight model in existence for laptops with integrated graphics. At 2.3GB VRAM it runs on practically any machine at 70+ tok/s for zero dollars.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Extract all action items, assignees, and deadlines from this meeting transcript into JSON.

PREVIEW NOTE

Valid JSON schema with cleanly parsed tasks and dates.

950

Hugging Face↗

●LOCAL • 82/100

Wan 2.1 14B

Wan / Alibaba

Wan-AI's official 14B open-weight video foundation model for text-to-video and image-to-video generation, with native ComfyUI and Comfy-Org modular workflow repacks.

BASE ARCHITECTURE / Wan 2.1 14B (T2V / I2V Diffusion Transformer)

SRuns well (100/100) • GGUF Q4 (~118 tok/s • 18GB free)

ℹTHE BRUTAL TRUTH

The most accessible open video foundation model released to date. Apache 2.0 with modular Comfy-Org safetensors and low-VRAM GGUF builds.

No cardLocalComfyUIComfy-Org RepackOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

A slow cinematic tracking shot across a rain-soaked neon street.

PREVIEW NOTE

Photorealistic 720p cinematic video sequence.

Original Author:Wan-AI/Wan2.1-T2V-14B ↗

Comfy-Org Repack:Comfy-Org/Wan_2.1_ComfyUI_repackaged ↗

2450k

Comfy-Org Hugging Face↗

●LOCAL • 95/100

Whisper Large v3 Turbo

OpenAI

OpenAI's latest optimized automatic speech recognition (ASR) model. Delivers 8x faster transcription with near-zero word error rate under MIT license.

BASE ARCHITECTURE / Whisper Large v3 Turbo

SRuns well (97/100) • FP16 (~176 tok/s • 20GB free)

ℹTHE BRUTAL TRUTH

The absolute undisputed king of speech-to-text. 8x faster than before, runs in 4GB VRAM, and carries a pure MIT license. Stop paying for Whisper APIs.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Transcribe an hour-long multi-speaker meeting with timestamps and punctuation.

PREVIEW NOTE

A precise JSON transcript with segment timestamps, speaker pauses, and accurate casing.

910

Hugging Face↗

●LOCAL • 89/100

Qwen2.5-Coder-7B-Instruct

Alibaba / Qwen

State-of-the-art 7B code LLM rivaling older 34B models. Fits on virtually any 6GB–8GB GPU with 128K context and Apache 2.0 licensing.

BASE ARCHITECTURE / Qwen2.5-Coder-7B-Instruct

SRuns well (98/100) • Q4_K_M (~141 tok/s • 19GB free)

ℹTHE BRUTAL TRUTH

The absolute king of 8GB GPUs for coding. It beats models four times its size, runs at 80 tok/s on an RTX 3060, and carries Apache 2.0.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Write a SQL migration script to add soft deletes and index created_at on user tables.

PREVIEW NOTE

A PostgreSQL migration script adding deleted_at TIMESTAMP WITH TIME ZONE and a partial index.

840

Hugging Face↗

●LOCAL • 87/100

Hermes 3 (Llama 3.1 8B)

Nous Research

State-of-the-art generalist instruction and roleplay model by Nous Research. Superb function calling, code generation, and multi-turn planning.

BASE ARCHITECTURE / Hermes-3-Llama-3.1-8B

SRuns well (98/100) • Q4_K_M (~141 tok/s • 19GB free)

ℹTHE BRUTAL TRUTH

The best community fine-tune of Llama 3.1 8B. Highly steerable, great at structured tool calling, and free of annoying canned refusals.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Plan a 5-step automated penetration test checklist for an OAuth 2.0 authorization code flow with PKCE.

PREVIEW NOTE

A tactical security audit checklist testing token leakage, replay attacks, and state CSRF validation.

760

Hugging Face↗

●LOCAL • 83/100

SmolLM2-1.7B-Instruct

Hugging Face

Ultra-compact on-device language model trained on 11 trillion tokens by Hugging Face. Runs at 120+ tokens/sec on CPU or 2GB VRAM with Apache 2.0.

BASE ARCHITECTURE / SmolLM2-1.7B-Instruct

SRuns well (94/100) • Q4_K_M (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

The best sub-2B model in the world. Ideal for edge IoT, Raspberry Pis, and instant regex/JSON extraction at 120+ tok/s on CPU.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Extract customer name, order ID, and refund reason into clean JSON from this email.

PREVIEW NOTE

A structured JSON object with parsed customer fields and zero markdown conversational filler.

710

Hugging Face↗

●LOCAL • 88/100

LTX-Video (2B Realtime)

Lightricks

Real-time 2B spatial-temporal video DiT capable of generating 24fps high-resolution video in under 15 seconds on consumer GPUs under Apache 2.0.

BASE ARCHITECTURE / LTX-Video 0.9.1 (2B)

ARuns well (79/100) • BF16 (~78 tok/s • 15GB free)

ℹTHE BRUTAL TRUTH

Generates 24fps video in seconds instead of minutes. On an 8GB card, it feels like magic compared to the painful wait times of older video models.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

A golden retriever running through autumn leaves in slow motion.

PREVIEW NOTE

A 24fps video sequence rendered in under 18 seconds with smooth fur movement and flying leaves.

710

Hugging Face↗

●LOCAL • 89/100

Ministral-8B-Instruct

Mistral AI

Mistral's edge-computing champion with a 128K context window and interleaved sliding window attention, designed for sub-second agentic decision loops.

BASE ARCHITECTURE / Ministral-8B-Instruct-2410

SRuns well (98/100) • Q4_K_M (~141 tok/s • 19GB free)

ℹTHE BRUTAL TRUTH

Phenomenal edge latency and tool calling accuracy, but restricted by Mistral's research license for production commercial SaaS.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Decide whether to execute tool get_weather or search_flights based on query: 'Flight to Boston tomorrow morning'.

PREVIEW NOTE

A structured tool-call payload targeting search_flights with destination 'BOS' and departure date.

705

Hugging Face↗

●LOCAL • 99/100

Liquid LFM3 1.7B

Liquid AI

Liquid AI's Liquid Foundation Model (LFM3) built from first principles with gated convolutions and dynamical systems. Delivers frontier-grade tool calling and agent reasoning with a 256K context at minimal memory footprint.

BASE ARCHITECTURE / LFM3-1.7B (Liquid Dynamical Architecture)

SRuns well (94/100) • Q4_K_M (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

The most promising post-transformer architecture on the market. If you are building local agent loops that call tools continuously, LFM3 gives you 120+ tokens/sec on almost no VRAM with a 256K context.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Coordinate a 3-agent research synthesis with structured tool schema execution.

PREVIEW NOTE

Accurate function calling payload with verified parameter validation.

680

Hugging Face↗

●LOCAL • 84/100

Kokoro-82M

Kokoro / community

A compact 82M text-to-speech model suited to local audio pipelines, CPU inference, and lightweight browser prototypes.

BASE ARCHITECTURE / Kokoro-82M

SRuns well (95/100) • ONNX CPU (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

Runs on CPU, tiny footprint, Apache 2.0 — the only reason not to use it is voice quality. If you need premium voices, pay for ElevenLabs.

No cardLocalOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Welcome back. Your next idea starts here.

PREVIEW NOTE

Illustrative speech text; no audio is generated by this page.

632

Hugging Face↗

●LOCAL • 88/100

MOSS Image Base

OpenMOSS Team

User guide   ·   中文用户指南   ·   ComfyUI

BASE ARCHITECTURE / MOSS Image Base (5.1B Flow-Matching Diffusion Transformer)

SRuns well (97/100) • Q2_K (2-bit Extreme) (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

Audited in Stactora CMS. Architecture verified with 5.1B parameters.

No cardLocalComfyUIComfy-Org RepackOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

A wide view of a refrigerator beside a kitchen counter with a fruit bowl and a sticky note.

PREVIEW NOTE

Photorealistic generative composition with clear embedded typography.

Original Author:OpenMOSS-Team/MOSS-Image-Base ↗

Comfy-Org Repack:Comfy-Org/MOSS-Image ↗

168k

Comfy-Org Hugging Face↗

●LOCAL • 85/100

Granite-3.0-8B-Instruct

IBM Research

IBM's flagship enterprise open-weight model with 128K context, trained under strict copyright curation and released with full Apache 2.0 licensing and IP indemnity.

BASE ARCHITECTURE / Granite-3.0-8B-Instruct

SRuns well (98/100) • Q4_K_M (~141 tok/s • 19GB free)

ℹTHE BRUTAL TRUTH

Built specifically for enterprise legal departments terrified of training data copyright lawsuits. Solid RAG performance with true Apache 2.0.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Draft an enterprise compliance policy for data classification across Confidential, Internal, and Public.

PREVIEW NOTE

A formal compliance document with handling procedures, access controls, and retention schedules.

530

Hugging Face↗

●LOCAL • 98/100

Audio8-TTS Realistic 0.6B

Edge0

Edge0's ultra-realistic 601M parameter open-weight Text-to-Speech and zero-shot voice cloning model. Bundles an integrated 44.1kHz neural audio codec and 23-language multilingual synthesis under Apache 2.0.

BASE ARCHITECTURE / Audio8-TTS-Realistic-0.6b (DualAR Architecture)

SRuns well (92/100) • INT8 (~220 tok/s • 23GB free)

ℹTHE BRUTAL TRUTH

One of the cleanest sub-1B TTS engines in the open-source ecosystem. Bundles its own 44.1kHz neural codec, clones voices in 3 seconds, and runs on literally any 2GB laptop GPU under Apache 2.0.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Clone this 5-second speaker reference and synthesize technical documentation narration.

PREVIEW NOTE

Studio-grade 44.1kHz stereo audio stream with accurate speaker timbre and cadence.

510

Hugging Face↗

●LOCAL • 98/100

Spark X2.6 1.7B

SparkLLM Team / XHToken

XHToken's state-of-the-art 1.7B on-device reasoning and coding model. Features a 1M token context length, configurable thinking effort modes (none/low/high), and outsized performance on SWE-bench and autonomous planning.

BASE ARCHITECTURE / Spark-X2.6-1.7B (Sliding-Attention Transformer)

SRuns well (94/100) • Q4_K_M (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

The undisputed champion of 1.7B on-device reasoning. Having a controllable thinking budget (none/low/high) plus a 1M token context under Apache 2.0 makes it the best lightweight agent brain available.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Analyze this 20-file repo trace, identify the race condition, and output a minimal patch.

PREVIEW NOTE

Step-by-step thinking trace followed by precise diff patch.

460

Hugging Face↗

●LOCAL • 91/100

IDOL (Image-to-3D DiT)

Speridlabs

Speridlabs' open-source 3D Diffusion Transformer (3D-DiT). Converts single photos into dense 3D Gaussian Splats (up to 1,048,576 Gaussians) outputting standard .ply files under an Apache 2.0 license.

BASE ARCHITECTURE / IDOL 3D-DiT (0.66B Generator / 1.8B Pipeline)

SRuns well (92/100) • FP8 (~123 tok/s • 18GB free)

ℹTHE BRUTAL TRUTH

The most practical open-weight 3D Gaussian generator today. Requiring only ~7.2 GB VRAM under Apache 2.0 makes high-fidelity 3D spatial generation accessible on standard laptops and desktops.