Models

Results

• LElerobot/smolvla_base
  Vision-language-action450M paramsEst. Edge GPUApache 2.0Jun 2025

  A compact 450M-parameter VLA from Hugging Face, pretrained on community-collected LeRobot datasets and sized for fine-tuning and inference on consumer GPUs.

• LElerobot/pi05_base
  Vision-language-action3.6B paramsEst. 24 GB GPUGemma termsSep 2025

  LeRobot's port of the Physical Intelligence pi0.5 base model, which is co-trained on web and multi-robot data to generalize better to unfamiliar homes and objects.

• NVnvidia/GR00T-N1.6-3B
  Vision-language-action3.3B paramsEst. Edge GPUNVIDIA One-Way Noncommercial LicenseDec 2025

  A 3B cross-embodiment robot foundation model from NVIDIA that combines a vision-language backbone with a diffusion-transformer action head and can be post-trained on a new robot with a modest amount of data.

• OPopenvla/openvla-7b
  Vision-language-action7.5B paramsEst. 24 GB GPUMITJun 2024

  A 7B vision-language-action model on a Llama 2 backbone that maps one camera image and a text instruction to discretized 7-DoF end-effector actions, trained on Open X-Embodiment episodes.

• RBrail-berkeley/octo-base-1.5
  Vision-language-action93M paramsEst. Edge GPUMITMay 2024

  A 93M-parameter transformer generalist policy with a diffusion action head, conditioned on language or goal images and intended to be fine-tuned to new robots, sensors, and action spaces.

• LElerobot/pi0_base
  Vision-language-action3.5B paramsEst. 24 GB GPUGemma termsSep 2025

  LeRobot's port of the Physical Intelligence pi0 base checkpoint, a PaliGemma-based VLA that outputs continuous action chunks through a flow-matching action expert.

• LElerobot/pi0fast-base
  Vision-language-action2.9B paramsEst. 24 GB GPUGemma termsJan 2026

  An autoregressive pi0 variant that predicts actions as compressed FAST tokens from a DCT-based tokenizer instead of using flow matching, packaged for LeRobot.

• RDrobotics-diffusion-transformer/rdt-1b
  Vision-language-action1.2B paramsEst. 24 GB GPUMITAug 2024

  A roughly 1B-parameter diffusion transformer policy that predicts bimanual action chunks in a unified action space shared across many robots, conditioned on language and up to three camera views.

• COCogACT/CogACT-Base
  Vision-language-action7.6B paramsEst. 24 GB GPUMITNov 2024

  A VLA that uses a 7B vision-language model for high-level understanding and passes its output to a separate diffusion-transformer module that generates continuous action sequences.

• LElerobot/act_aloha_sim_transfer_cube_human
  Imitation policy52M paramsEst. Edge GPUApache 2.0May 2024

  An Action Chunking Transformer policy trained in LeRobot on human demonstrations of the simulated ALOHA bimanual cube-transfer task.

• LElerobot/diffusion_pusht
  Imitation policy263M paramsEst. Edge GPUApache 2.0May 2024

  A Diffusion Policy checkpoint trained in LeRobot to push a T-shaped block into a target pose in the 2D PushT simulation.

• NVnvidia/Cosmos-Predict2.5-2B
  World model2B paramsEst. DatacenterNVIDIA Open Model LicenseGatedSep 2025

  A 2B-parameter video world model that generates future video from text, image, or video prompts, with extra multiview checkpoints for robot and driving simulation.

• FAfacebook/vjepa2-vitl-fpc64-256
  Vision encoder326M paramsEst. Edge GPUMITMay 2025

  Meta's self-supervised ViT-L video encoder, pretrained by predicting masked video regions in latent space rather than in pixels.

• NVnvidia/Cosmos-Reason2-2B
  Vision-language2.4B paramsEst. 24 GB GPUNVIDIA Open Model LicenseGatedDec 2025

  A small reasoning vision-language model fine-tuned from Qwen3-VL-2B for physical common sense, such as judging object interactions, spatial relations, and whether a video is physically plausible.

• KYkyutai/moshiko-pytorch-bf16
  Speech to speech7.7B paramsEst. 24 GB GPUCC BY 4.0Sep 2024

  Kyutai's full-duplex speech-to-speech foundation model in its male-voice variant, which models the user's and its own speech as parallel streams and reaches about 200 ms latency in practice.

• FAfixie-ai/ultravox-v0_5-llama-3_2-1b
  Speech to textEst. Edge GPUMITFeb 2025

  A speech-input LLM that feeds Whisper encoder features through a learned projector into Llama 3.2 1B Instruct, so it can respond to spoken input without a separate transcription step. It replies in text, so pair it with a TTS voice.

• QWQwen/Qwen3-Omni-30B-A3B-Instruct
  Speech to speech35B paramsEst. DatacenterApache 2.0Sep 2025

  A mixture-of-experts omni model with about 3B active parameters that takes text, image, audio, and video as input and replies in both text and streaming speech.

• OPopenai/whisper-large-v3-turbo
  Speech to text809M paramsEst. Edge GPUMITOct 2024

  A faster Whisper large-v3 derivative whose decoder is cut from 32 layers to 4, giving multilingual transcription in roughly 100 languages at a small accuracy cost (it was not fine-tuned for translation).

• MAmoonshine-ai/moonshine-streaming-small
  Speech to text123M paramsEst. Single-board computerMITJan 2026

  The small size of the Moonshine Streaming family, an English-only encoder-decoder recognizer that uses a sliding-window encoder so it can transcribe live audio with low latency on edge hardware.

• NVnvidia/parakeet-tdt-0.6b-v3
  Speech to text627M paramsEst. Edge GPUCC BY 4.0Aug 2025

  A FastConformer encoder with a token-and-duration transducer decoder that transcribes 25 European languages with punctuation, capitalization and word-level timestamps.

• NVnvidia/nemotron-3.5-asr-streaming-0.6b
  Speech to text638M paramsEst. Edge GPUOpenMDW 1.1May 2026

  A cache-aware FastConformer-RNNT streaming recognizer that covers 40 language-locales from one checkpoint via language-ID prompting, outputs punctuated text, and lets you pick chunk sizes from 80 ms up to about 1.1 s.

• KYkyutai/stt-1b-en_fr
  Speech to text989M paramsEst. Edge GPUCC BY 4.0Jun 2025

  A decoder-only streaming English and French transcriber based on Kyutai's delayed-streams design, running about 0.5 s behind the audio and exposing a semantic voice-activity signal for end-of-turn detection.

• GGggerganov/whisper.cpp
  Speech to textEst. Single-board computerMITMar 2023

  GGML-format conversions of the OpenAI Whisper checkpoints, from tiny to large, for the whisper.cpp C/C++ inference engine.

  Works with GotchiGotchi Local can download this as an offline speech fallback on your computer.
• HEhexgrad/Kokoro-82M
  Text to speech82M paramsEst. Single-board computerApache 2.0Dec 2024

  An 82M-parameter text-to-speech model built on StyleTTS 2 components, shipped with a set of preset voices across several languages.

• RHrhasspy/piper-voices
  Text to speechEst. Single-board computerMITJun 2023

  The official collection of ONNX voice models for the Piper offline TTS engine, covering many languages at several size and quality levels.

  Works with GotchiGotchi Local can download this as an offline speech fallback on your computer.
• KYkyutai/pocket-tts
  Text to speech100M paramsEst. Single-board computerCC BY 4.0GatedDec 2025

  A 100M-parameter CPU-oriented TTS model from Kyutai with voice cloning from a reference clip and support for English plus five other European languages, designed to run faster than real time using two CPU cores.

• SEsesame/csm-1b
  Text to speech1.55B paramsEst. Edge GPUApache 2.0GatedMar 2025

  A conversational speech model that produces Mimi audio codes from text and prior dialogue audio, so its prosody follows the flow of the conversation.

• REResembleAI/chatterbox
  Text to speech500M paramsEst. Edge GPUMITApr 2025

  A family of 0.5B zero-shot voice-cloning TTS models whose repo holds the English model, which has an emotion-exaggeration control, alongside multilingual checkpoints covering 23 languages.

• QWQwen/Qwen3-TTS-12Hz-0.6B-Base
  Text to speech915M paramsEst. Edge GPUApache 2.0Jan 2026

  The smaller base checkpoint of Qwen3-TTS, a streaming text-to-speech model on a 12 Hz speech tokenizer that covers 10 languages and clones a voice from a roughly 3-second reference clip.

• CAcanopylabs/orpheus-3b-0.1-ft
  Text to speech3.8B paramsEst. LaptopApache 2.0GatedMar 2025

  A Llama-based speech LLM fine-tuned for English TTS with preset voices, inline non-verbal tags such as laughs and sighs, and streaming output.

• NLnari-labs/Dia-1.6B
  Text to speech1.6B paramsEst. LaptopApache 2.0Apr 2025

  An English-only dialogue TTS model that turns a two-speaker script into conversation audio, including non-verbal sounds such as laughing or coughing.

• OConnx-community/silero-vad
  Voice activity & turnsEst. Single-board computerMITJul 2024

  ONNX exports of the Silero voice activity detector in fp32, fp16 and 8-bit forms (the int8 file is about 0.6 MB) that mark which short audio frames contain speech.

• PApipecat-ai/smart-turn-v3
  Voice activity & turns8M paramsEst. Single-board computerBSD 2-ClauseSep 2025

  An 8M-parameter multilingual turn-detection model, a Whisper Tiny encoder with a linear head, that predicts from raw audio whether the speaker has finished; the int8 CPU ONNX build is about 8.7 MB.

• DAdavidscripka/openwakeword
  Wake wordEst. Single-board computerCC BY-NC-SA 4.0Oct 2023

  Pretrained openWakeWord models, small classifiers on top of a shared speech-embedding front end, that detect phrases such as 'hey jarvis' in a continuous audio stream.

• PYpyannote/speaker-diarization-community-1
  Speaker diarizationEst. Single-board computerCC BY 4.0GatedSep 2025

  The open speaker-diarization pipeline released with pyannote.audio 4.0, combining a segmentation model, speaker embeddings and PLDA-based clustering to label who spoke when, with better speaker counting than the 3.1 pipeline.

• FAfacebook/sam2.1-hiera-small
  Segmentation46M paramsEst. Edge GPUApache 2.0Sep 2024

  Compact variant of Meta's promptable segmentation model that returns object masks from points or boxes and can track them across video frames.

• FAfacebook/sam3
  Segmentation860M paramsEst. 24 GB GPUCustom licenseGatedNov 2025

  Meta's third-generation segmentation model that can find, mask and track every instance of a concept described by a short text phrase or example image.

• IRIDEA-Research/grounding-dino-tiny
  Object detection172M paramsEst. Edge GPUApache 2.0Sep 2023

  Open-vocabulary detector that draws boxes around objects named in a free-text prompt, without a fixed class list.

• GOgoogle/owlv2-base-patch16-ensemble
  Object detection155M paramsEst. Edge GPUApache 2.0Oct 2023

  CLIP-based zero-shot detector from Google that localizes objects from text queries or from an example image crop.

• UCustc-community/dfine-small-coco
  Object detection10M paramsEst. Single-board computerApache 2.0Feb 2025

  Small real-time DETR-style detector for the 80 COCO classes that refines box edges as distributions for tighter localization.

• DAdepth-anything/Depth-Anything-V2-Small-hf
  Depth estimation25M paramsEst. Single-board computerApache 2.0Jun 2024

  Lightweight monocular model that predicts a relative depth map from a single RGB frame.

• APapple/DepthPro-hf
  Depth estimation952M paramsEst. 24 GB GPUApple ML ResearchNov 2024

  Apple's monocular model that outputs sharp metric depth plus an estimated focal length from one image, with no camera intrinsics needed.

• UCusyd-community/vitpose-base-simple
  Pose & keypoints86M paramsEst. Edge GPUApache 2.0Jan 2025

  Plain vision-transformer model that estimates 2D human body keypoints for each person box produced by a separate detector.

• ECETH-CVG/lightglue_superpoint
  Pose & keypoints14M paramsEst. Edge GPUCustom licenseFeb 2025

  Learned matcher that pairs SuperPoint keypoints between two images and stops early on easy pairs to save compute.

• FAfacebook/dinov3-vits16-pretrain-lvd1689m
  Vision encoder22M paramsEst. Single-board computerDINOv3 LicenseGatedAug 2025

  Smallest ViT from Meta's DINOv3 family, a self-supervised backbone whose dense patch features transfer to many vision tasks without fine-tuning.

• HUHuggingFaceTB/SmolVLM2-500M-Video-Instruct
  Vision-language507M paramsEst. Single-board computerApache 2.0Feb 2025

  Half-billion-parameter vision-language model from Hugging Face that answers questions about images and short video clips.

• VIvikhyatk/moondream2
  Vision-language1.9B paramsEst. Edge GPUApache 2.0Mar 2024

  Compact, frequently updated vision-language model with built-in captioning, visual Q&A, object detection and pointing.

• QWQwen/Qwen3-VL-2B-Instruct
  Vision-language2.1B paramsEst. Edge GPUApache 2.0Oct 2025

  Smallest instruction-tuned model in Alibaba's Qwen3-VL line, covering image and video understanding, OCR and grounded object localization.

• GOgoogle/paligemma2-3b-mix-224
  Vision-language3B paramsEst. Edge GPUGemma termsGatedNov 2024

  Google's 3B vision-language model trained on a mix of tasks, including captioning, visual Q&A, detection and segmentation, at 224px input.

• GOgoogle/gemma-3n-E2B-it
  Vision-language5.4B paramsEst. Edge GPUGemma termsGatedJun 2025

  Mobile-first Gemma variant that accepts text, image and audio input and runs with a memory footprint close to a 2B model by offloading per-layer embeddings.

• HUHuggingFaceTB/SmolLM3-3B
  Small language model3.1B paramsEst. Edge GPUApache 2.0Jul 2025

  Fully open 3B language model with switchable reasoning mode, long context, tool calling and multilingual support, released with its training recipe.

• QWQwen/Qwen3-0.6B
  Small language model751M paramsEst. Single-board computerApache 2.0Apr 2025

  Smallest dense model in the Qwen3 family, with optional thinking mode and function-calling support.

• GOgoogle/gemma-3-270m-it
  Small language model270M paramsEst. Single-board computerGemma termsGatedJul 2025

  Tiny instruction-tuned Gemma 3 model meant to be fine-tuned for narrow jobs such as classification, extraction or command routing.