Model reference

Compact LLM optimized for step-by-step reasoning and logic tasks.

English ASR model with low error rates, trained on LibriSpeech data.

Fast, compact model for English speech recognition and transcription

Ultra-light instruction-tuned LLM for fast, low-cost text tasks.

High-quality text-to-image model—slower than Schnell, sharper than SDXL.Spec sheet. 12 billion-parameter rectified-flow transformer, trained with guidance distillation; works best at 20-40 denoise steps for 1024 × 1024 frames…

Open-weight, text-guided image editor that keeps your characters and style intact. One-prompt edits, zero drift. Specify a change (“add a hat,” “swap background”) and the rectified-flow transformer applies it while preserving character, style and object details—even across multiple successive tweaks…

Fast-track text-to-image model distilled for speed.1-to-4-step renders. Rectified-flow backbone + latent adversarial distillation means high-quality 1024-px frames in as few as four denoise loops (one if you crank the speed)…

Tiny transformer that stays within 8 k tokens yet delivers solid reasoning.Spec sheet. Decoder-only, RoPE positional encodings, 8 192-token context window; pre-trained on ~2 T tokens of web, code & math text…

Open-source workhorse that still fits on a single mid-range card.Spec sheet. 28 decoder layers, 4 096-dim model, 16 heads, RoPE, 2 048-token window.Trained on The Pile. 400 B mixed-domain tokens give it solid general knowledge and code skills…

Ultra-light LLM for fast, safe, large-scale text processing.

High-end text-to-video focused on cinematic motion and scene depth.

Multimodal model for unified understanding and image generation.

SDXL-quality images at SSD-speed.Big images, blink-fast. Spits out a 1024 × 1024 frame in ≈ 0 .66 s with 10 denoise steps on an RTX 4090—about 4 × quicker than plain SDXL…

Compact Meta transformer that packs 128 k-token memory into an 8 GB GPU budget.Long-context, low-footprint. Full-precision model reads up to 128 k tokens with GQA; BF16 weights clock in at ≈7.4 GB VRAM, while 4-bit SpinQuant/QLoRA builds shrink below 4 GB for laptops and edge devices…

Fast, controllable text-to-video for quick iteration on motion and scenes.

Efficient 3B instruction-tuned LLM for fast, reliable text tasks.

Open-science transformer you can run on a laptop.Spec sheet. 16-layer decoder-only stack, 2 048 hidden size, 16 attention heads, 4 096-token window; pre-trained on 4 T tokens of mixed web/code text…

Open-source text-to-speech model with customizable voices.

Pocket-sized transformer that thinks like a bigger model.Spec sheet. 32-layer decoder-only network, 2 048-token window, trained on 1.4 T “textbook-quality” tokens of filtered web + synthetic data…

Instruction‑tuned small language model with strong reasoning and a longer context window.

Compact instruction-tuned LLM with strong reasoning for its size.

SDXL-class 1024-px images in four diffusion steps.1024-px in one shot. Pure-Transformer latent model spits out 1024 × 1024 frames straight from the prompt—no upscaler or second pass needed…

EleutherAI’s open, research-first transformer; small enough for one card, good enough to beat OPT/GPT-Neo peers. Spec sheet. 32 layers · 2 560 hidden size · 32 heads · 2 048-token context; trained on The Pile with the GPT-NeoX codebase…

Pocket-sized Qwen that still handles marathon prompts.Spec sheet. 28-layer decoder with Grouped-Query Attention (12 Q / 2 KV heads), RoPE, SwiGLU. Native 32 k-token window; stretch to 128 k with YARN/Dual-Chunk tricks…

Small-footprint transformer that punches above its weight.Dual-mode reasoning. Toggle enable_thinking to get full chain-of-thought for hard math/code problems, or turn it off for fast, lightweight chat. One model, two personas…

Next-gen text-to-image with stronger prompt alignment and cleaner visuals.

Near-real-time text-to-image generation with minimal diffusion steps.

Little-footprint Llama-style model that hits well above its weight.Spec sheet. 1.7 B-param decoder, 8 192-token context, trained on 11 T tokens (FineWeb-Edu, DCLM, The Stack + new math/code sets)…

Text-to-song generation focused on full musical structure, not clips.

Latent-diffusion upgrade that trades the old CLIP-L text encoder for a beefier OpenCLIP-ViT/H and cleaner training data.Sharper prompt understanding. New OpenCLIP-H embedding makes prompts more literal and gives you better edge detail than v1.5—just note that old prompt hacks won’t map 1-to-1…

Text-to-image generator that fits on a single consumer GPU.What it is. Latent-diffusion model that turns plain-English prompts into 512 × 512 images; upscale or edit after if you need bigger…

Pocket-size Llama-2 drop-in you can run almost anywhere.4 K-token brain. Keeps Llama-2’s tokenizer and 4 096-token RoPE window, so existing prompts and LoRA adapters just work…

High-quality image-to-video generation focused on smooth, consistent motion.

High-fidelity text-to-video focused on stable scenes and motion.

Text + image guided video generation with balanced quality and compute.

English speech recognition model trained on 960 hours of audio.

Advanced model for speech recognition and transcription.

Multilingual model for speech recognition and audio transcription.

Multilingual voice cloning model—clone any voice in 17 languages from a short sample.