Modelos de IA que utilizamos - Todos de código abierto Free.ai

Apache 2.0 24B Self-hosted

High-quality 7B model from Mistral AI with excellent instruction following.

by Mistral AI

Mistral Small 3 (24B)

Mistral Small 3 — Apache 2.0. 24B dense, fast inference, strong multilingual. Drop-in replacement …

by Mistral AI

Phi-4 14B

MIT 14B Self-hosted

Phi-4 — MIT licensed. Microsoft's strongest small model, trained on curated synthetic data.

Apache 2.0 32B Self-hosted

Qwen3 32B

Mid-size Qwen3. Apache 2.0. Strong multilingual + reasoning. Fits on a single 20GB A100.

by Alibaba

Qwen3 Next 80B

Apache 2.0 80B Self-hosted

Qwen3-Next-80B is Alibaba's flagship open reasoning model released late 2026. Apache 2.0 licensed. Competitive …

by Alibaba

SmolLM 3 3B

Apache 2.0 3B Self-hosted

SmolLM 3 — Apache 2.0, Hugging Face's tiny-but-capable open model. Edge deployment friendly.

by HuggingFace

Explore more chat models

Apache 2.0 32B Self-hosted

Code Generation 3

Qwen 2.5 Coder 32B

Best open-source code model. Excels at Python, JavaScript, TypeScript, and 50+ languages.

Qwen3-Coder 7B

Dedicated code model optimized for programming tasks. Supports 100+ languages.

BigCode OpenRAIL-M 15B Self-hosted

StarCoder 2 15B

StarCoder2 — OpenRAIL-M (commercial-allowed with use-restrictions). Trained on permissively-licensed GitHub.

by BigCode

Explore more code models

Apache 2.0 12B Self-hosted

Image Generation 8

FLUX.1 Krea

FLUX.1 Krea — Apache 2.0 open variant of FLUX tuned for photorealistic output.

by Black Forest Labs

FLUX.1 Schnell

Apache 2.0 12B Self-hosted

State-of-the-art text-to-image model. Fast generation with stunning quality.

by Black Forest Labs

FLUX.2 Klein 4B

Apache 2.0 4B Self-hosted

Latest FLUX image generation model with improved quality and style diversity.

by Black Forest Labs

Kolors

Apache 2.0 2.5B Self-hosted

Kolors — Apache 2.0. ChatGLM3-based text encoder; exceptional Chinese-English bilingual prompts.

by Kuaishou

Lumina Image 2

Apache 2.0 2.6B Self-hosted

Lumina Image 2 — Apache 2.0. DiT-based, strong prompt following, highly controllable.

by Alpha-VLLM

PixArt-Σ

OpenRAIL++ 600M Self-hosted

PixArt-Σ — OpenRAIL++. Transformer-based 4K-capable image gen at a fraction of SDXL's compute.

by PixArt-alpha

SDXL Turbo

OpenRAIL++ 3.5B Self-hosted

SDXL Turbo — OpenRAIL++. Single-step image generation; order-of-magnitude faster than SDXL.

by StabilityAI

Stable Diffusion XL

OpenRAIL++ 3.5B Self-hosted

Industry-standard image generation model with wide community support.

by Stability AI

Explore more image models

Apache 2.0 1B adapter Self-hosted

Video Generation 6

AnimateDiff v3

AnimateDiff — Apache 2.0. Motion adapter that animates SDXL/SD 1.5 images.

by AnimateDiff

CogVideoX-2B

Apache 2.0 2B Self-hosted

Text-to-video generation model producing 4-8 second clips.

by Tsinghua University

HunyuanVideo 1.5

Apache 2.0 1.5B Self-hosted

Generate stunning videos from text prompts. Higher quality than CogVideoX with more natural motion.

by Tencent

Mochi 1

Apache 2.0 10B Self-hosted

Mochi 1 — Apache 2.0! The only fully-open frontier text-to-video model. Strong motion, cinematic.

by Genmo

SadTalker

MIT 500M Self-hosted

SadTalker — MIT. Audio-driven talking-head video generation. Photo + audio in, lip-synced video out. …

by OpenTalker

Wan 2.1

Apache 2.0 14B Self-hosted

Wan 2.1 — Apache 2.0! First fully-open competitive T2V+I2V model.

by Alibaba

Explore more video models

Text-to-Speech (TTS) 9

Bark

MIT 1.5B Self-hosted

Bark — MIT licensed text-to-audio model that generates speech, music, sound effects, and non-speech …

by Suno AI

Chatterbox

Highly expressive TTS with voice cloning support.

by Resemble AI

Kokoro

Apache 2.0 Self-hosted

Ultra-fast TTS with natural intonation. Supports streaming.

by Hexgrad

MeloTTS

Apache 2.0 2.3B Self-hosted

High-quality multilingual text-to-speech model.

by MyShell

OuteTTS

Apache 2.0 Self-hosted

TTS with built-in voice cloning from short audio samples.

by Oute

Parler-TTS Large

Parler-TTS — Apache 2.0. Fully open reproduction of Google's Parler. Control tone, pace, and …

by HuggingFace

Piper

Apache 2.0 0.6B Self-hosted

Fast and lightweight TTS with dozens of voices and languages.

by Rhasspy

Qwen3-TTS 0.6B

Advanced TTS with 3-second voice cloning. Clone any voice from a short sample. Multilingual …

StyleTTS 2

MIT 24M Self-hosted

StyleTTS 2 — MIT licensed, zero-shot voice cloning from a single reference sample. Small, …

by Yinghao Aaron Li

Explore more TTS models

Speech-to-Text (STT) 3

Distil-Whisper large-v3

MIT 756M Self-hosted

Distil-Whisper — MIT licensed. 6× faster than Whisper large-v3 with ~1% WER gap. Drop-in …

by HuggingFace

Moonshine Base

MIT 61M Self-hosted

Moonshine — MIT. Optimized for edge devices; faster than Whisper tiny at better accuracy.

by Useful Sensors

faster-whisper large-v3

CTranslate2 port of OpenAI Whisper. Fast and accurate speech recognition in 99 languages.

by OpenAI / SYSTRAN

Explore more STT models

Music Generation 2

ACE-Step 3.5B

MIT 3.5B Self-hosted

Generate complete songs with vocals, lyrics, and instrumentation in seconds. Supports any genre and …

by ACE-Step

AudioLDM 2

Apache 2.0 Self-hosted

Generate music from text descriptions. Any genre, mood, or style.

by Haohe Liu

Explore more music models

Apache 2.0 3B Self-hosted

Translation 2

MadLAD-400 3B

Google's Massive Multilingual Adaptive Dataset translation model. Supports 450+ languages with Apache 2.0 license.

by Google

Opus-MT

Apache 2.0 77M/pair Self-hosted

Opus-MT — Apache 2.0. 1000+ language-pair models. Tiny footprint, no licensing risk.

by Helsinki-NLP

Explore more translation models

Vision & Multimodal 3

InternVL 3 8B

MIT 8B Self-hosted

InternVL 3 — MIT. One of the strongest open VLMs, near Qwen-VL 72B quality …

by Shanghai AI Lab

Molmo 7B

Molmo — Apache 2.0 open vision-language model from AI2. Strong on document+chart understanding.

by Allen Institute

Qwen2.5-VL 7B

Advanced vision-language model that understands images, charts, and documents. Ask questions about any image.

Explore more vision models

OCR & Document Understanding 4

Florence-2 Large

MIT 770M Self-hosted

Florence-2 — MIT. Unified vision model doing OCR, object detection, captioning, and grounding in …

Apache 2.0 580M Self-hosted

GOT-OCR 2.0

Unified OCR model that handles text, tables, charts, and mathematical equations. Far superior to …

Apache 2.0 multi Self-hosted

PaddleOCR

PaddleOCR — Apache 2.0. Industrial-grade OCR supporting 80+ languages. Layout analysis, table detection, handwriting.

by PaddlePaddle

TrOCR Base

MIT 334M Self-hosted

TrOCR — MIT. Transformer-based OCR, especially strong on handwriting.

Explore more OCR models

Apache 2.0 333M Self-hosted

Image Enhancement & Upscaling 5

BiRefNet v2

MIT 220M Self-hosted

BiRefNet v2 — MIT. Bilateral reference dichotomous segmentation. Sharper than U-2-Net / RMBG on …

by ZhengPeng7

GFPGAN

GFPGAN — Apache 2.0. Face restoration + enhancement model. Fixes blurry/damaged portraits.

by Tencent

Marigold Depth

OpenRAIL++-M 865M Self-hosted

Marigold — OpenRAIL++-M. CVPR 2024 best paper monocular depth estimator, fine-tuned from Stable Diffusion. …

by ETH Zurich

Real-ESRGAN

BSD Self-hosted

State-of-the-art image upscaling and enhancement.

by Xintao Wang

Segment Anything 2

Apache 2.0 224M Self-hosted

SAM 2 — Apache 2.0. Meta's universal segmentation model, with video support.

by Meta

Explore more models

Audio Source Separation 2

Demucs

Separate vocals, drums, bass, and other stems from any song.

by Meta

Spleeter

MIT 20M Self-hosted

Spleeter — MIT. Stem separation (vocals, drums, bass, piano, other).

by Deezer

Explore more models

Apache 2.0 1B Self-hosted

3D Generation 3

InstantMesh

InstantMesh — Apache 2.0. Single-image → 3D mesh in seconds.

by TencentARC

TRELLIS

MIT 2B Self-hosted

TRELLIS — MIT. Unified 3D generation from text or image. Industry-leading mesh quality.

TripoSR

by VAST AI / Stability AI

Generate 3D meshes from single images in under a second. Outputs GLB or OBJ …

Explore more models

Text Embeddings 3

BGE-M3

Apache 2.0 305M Self-hosted

State-of-the-art multilingual text embedding model. Powers semantic search and similarity matching.

by BAAI

Nomic Embed v2

Nomic Embed v2 — Apache 2.0. Matryoshka-embeddings, truncatable 768→256→128 dims.

by Nomic AI

mxbai-embed-large-v1

Apache 2.0 335M Self-hosted

mxbai-embed-large-v1 — Apache 2.0. Strong on MTEB English benchmark.

by mixedbread.ai

Explore more models