Models

hh1-i2v

Alibaba's HappyHorse 1.0 image-to-video model. Animates a reference image with an optional text prompt. Supports 720P and 1080P output with durations from 3 to 15 seconds.

• Third-party
• Zero data retention

hh1-t2v

Alibaba's HappyHorse 1.0 text-to-video model. Generates videos from a text prompt with configurable resolution, aspect ratio, and duration (3-15s).

• Third-party
• Zero data retention

hh1.1-i2v

Alibaba's HappyHorse 1.1 image-to-video model. Animates a reference image with an optional text prompt, with smoother motion, natural skin textures, and improved close-up quality over 1.0. Supports 720P and 1080P output with durations from 3 to 15 seconds.

• Third-party

hh1.1-r2v

Alibaba's HappyHorse 1.1 reference-to-video model. Takes 1-9 reference images (characters and scenes) and a prompt that choreographs them into a single video, keeping each subject's identity consistent. Supports 720P and 1080P output with durations from 3 to 15 seconds.

• Third-party

hh1.1-t2v

Alibaba's HappyHorse 1.1 text-to-video model. Generates videos from a text prompt with stronger dynamic expressiveness, better visual quality, and improved instruction following over 1.0. Configurable resolution, aspect ratio, and duration (3-15s).

• Third-party

qwen3-max

Alibaba's Qwen 3 Max is a large language model with strong coding, reasoning, and multilingual capabilities, served via DashScope's OpenAI-compatible endpoint.

• Third-party
• Zero data retention

qwen3.5-397b-a17b

Alibaba's Qwen 3.5 is a 397B-parameter mixture-of-experts model with 17B active parameters, offering strong reasoning capabilities with efficient inference.

• Third-party
• Zero data retention

wan-2.6-image

Alibaba's Wan 2.6 text-to-image model generating images from text prompts with optional negative prompts and customizable dimensions.

• Third-party
• Zero data retention

wan-2.7-i2v

Alibaba's Wan 2.7 image-to-video model that generates videos from a reference image with optional text prompts. Supports 720P and 1080P output with durations from 2 to 15 seconds.

• Third-party
• Zero data retention

claude-fable-5

Claude Fable 5 is Anthropic's most capable widely released model, built for the most demanding reasoning and long-horizon agentic work. Adaptive thinking is always on, and the model supports a 1M token context window with up to 128k output tokens per request.

• Third-party

claude-haiku-4.5

Claude Haiku 4.5 delivers similar levels of coding performance at one-third the cost and more than twice the speed of larger models.

• Third-party
• Zero data retention

claude-opus-4.5

Claude Opus 4.5 brings further reasoning, coding, and agentic improvements over Opus 4.1, with stronger tool use and tighter instruction following.

• Third-party
• Zero data retention

claude-opus-4.6

Claude Opus 4.6 is Anthropic's flagship language model built for complex, multi-step work in coding, financial analysis, and legal reasoning. It uses extended thinking to work through complex problems carefully and features a one million token context window.

• Third-party
• Zero data retention

claude-opus-4.7

Claude Opus 4.7 is Anthropic's most capable generally available model, with a step-change improvement in agentic coding over Claude Opus 4.6. It uses adaptive thinking to calibrate reasoning per task and supports a one million token context window at standard pricing.

• Third-party
• Zero data retention

claude-opus-4.8

Claude Opus 4.8 is Anthropic's most capable generally available model, with a step-change improvement in agentic coding over Claude Opus 4.7. It uses adaptive thinking to calibrate reasoning per task and supports a one million token context window at standard pricing.

• Third-party
• Zero data retention

claude-opus-5

Claude Opus 5 is Anthropic's model for complex agentic coding and enterprise work, delivering intelligence close to Claude Fable 5 at half the price. It uses adaptive thinking to calibrate reasoning per task and supports a one million token context window at standard pricing.

• Third-party

claude-sonnet-4.5

Claude Sonnet 4.5 is the best coding model to date, with significant improvements across the entire development lifecycle.

• Third-party
• Zero data retention

claude-sonnet-4.6

Claude Sonnet 4.6 is Anthropic's latest balanced model offering strong coding, reasoning, and agentic capabilities with improved instruction following.

• Third-party
• Zero data retention

claude-sonnet-5

Claude Sonnet 5 is Anthropic's most agentic Sonnet model yet, built for coding, tool use, reasoning, and long-horizon professional work at lower cost than Opus-class models.

• Third-party

universal-3-pro

AssemblyAI's Universal 3 Pro speech recognition model for high-accuracy transcription.

• Third-party
• Zero data retention

flux-2-flex

FLUX.2 [flex] is Black Forest Labs' fine-grained control variant of FLUX.2 — exposes tunable inference steps, guidance, and prompt upsampling for typography-heavy and production workflows.

• Third-party

flux-2-max

FLUX.2 [max] is Black Forest Labs' highest-quality image model — top editing consistency, strongest prompt following, and grounding search for visualizations of real-time information.

• Third-party

flux-2-pro-preview

FLUX.2 [pro] Preview is Black Forest Labs' recommended default for production image generation and editing — tracks the latest [pro] weights with strong multi-reference support.

• Third-party

seedance-2.0

ByteDance's next-generation video model with a unified multimodal architecture. Generates high-quality video with synchronized audio from text, images, video clips, and audio inputs. Supports multimodal references (up to 9 images, 3 videos, 3 audio files), native audio generation, video editing, video extension, intelligent duration, and adaptive aspect ratio.

• Third-party

seedance-2.0-fast

Faster variant of ByteDance's Seedance 2.0 video model. Trades some quality for speed while sharing the same multimodal architecture. Supports text-to-video, image-to-video, native audio generation, multimodal references (images, videos, audio), video editing, and video extension.

• Third-party

seedance-2.0-mini

ByteDance's compact, cost-efficient video generation model from the Seedance 2.0 family. Supports text-to-video, image-to-video, reference video, and reference audio for background music. Ideal for high-volume workloads where speed and cost matter.

• Third-party

seedream-4.0

Seedream 4.0 is ByteDance's image creation model that combines text-to-image generation and image editing into a single architecture, offering fast, high-resolution output up to 4K.

• Third-party

seedream-4.5

Seedream 4.5 builds on 4.0 with multi-reference image support, batch generation, and sequential image generation.

• Third-party

seedream-5-lite

Seedream 5 Lite is a lighter, faster version of the Seedream 5 family with multi-reference and batch generation support.

• Third-party

seedream-5-pro

Seedream 5 Pro is ByteDance's high-quality image generation and editing model with text prompts, up to 10 reference images, and 1K, 2K, or explicit pixel-size output controls.

• Third-party

DeepSeek V4 Pro is a high-capability reasoning model from DeepSeek, served via Fireworks infrastructure for production-grade inference.

• Third-party

eleven-flash-v2-5

ElevenLabs' low-latency Flash v2.5 text-to-speech model for fast multilingual speech generation.

• Third-party

eleven-multilingual-v2

ElevenLabs' multilingual text-to-speech model for generating natural speech across many languages with ElevenLabs voices.

• Third-party

eleven-turbo-v2-5

ElevenLabs' Turbo v2.5 text-to-speech model balancing high-quality voice generation with low latency across 32 languages.

• Third-party

music-v2

ElevenLabs Music v2 composes songs and instrumental tracks from a prompt or detailed composition plan.

• Third-party

gemini-2.5-flash

Google's fast multimodal Gemini 2.5 model with strong reasoning and a 1M token context window.

• Third-party
• Zero data retention

gemini-2.5-flash-lite

Google's lightest and most cost-efficient Gemini 2.5 model for high-throughput tasks.

• Third-party
• Zero data retention

gemini-2.5-pro

Google's most capable Gemini 2.5 model with strong reasoning, thinking support, and a 1M token context window.

• Third-party
• Zero data retention

gemini-3-flash

Gemini 3 Flash is Google's fast multimodal model with frontier intelligence, superior search, and grounding capabilities.

• Third-party
• Zero data retention

gemini-3.1-flash-lite

Google's lightest and most cost-efficient Gemini model for high-throughput tasks.

• Third-party
• Zero data retention

gemini-3.1-flash-tts

• Third-party
• Zero data retention

gemini-3.1-pro

Google's most intelligent Gemini model with improved reasoning, a medium thinking level, and a 1M token context window.

• Third-party
• Zero data retention

gemini-3.5-flash

Gemini 3.5 Flash is Google's fast multimodal model with frontier intelligence, superior search, and grounding capabilities.

• Third-party

gemini-3.5-flash-lite

Gemini 3.5 Flash-Lite is a low-latency, cost-effective multimodal model optimized for high-throughput, low-cost execution for subagent tasks and document parsing.

• Third-party

gemini-3.6-flash

Gemini 3.6 Flash provides sustained frontier-level intelligence optimized for real-world tasks at a higher speed and lower cost, excelling at code generation, agentic execution, and spatial reasoning.

• Third-party

imagen-4

Google's latest image generation model producing high-quality, photorealistic images from text prompts with support for multiple aspect ratios.

• Third-party
• Zero data retention

nano-banana

Google's fast image generation model producing high-quality images from text prompts.

• Third-party
• Zero data retention

nano-banana-2

Google's second-generation image generation model with improved quality and speed.

• Third-party
• Zero data retention

nano-banana-2-lite

Google's fastest Gemini image generation model for rapid image creation and iteration.

• Third-party
• Zero data retention

nano-banana-pro

Google's higher-quality image generation model with improved detail and prompt adherence.

• Third-party
• Zero data retention

veo-3.1

Google's latest video generation model with improved quality, motion, and audio generation.

• Third-party
• Zero data retention

veo-3.1-fast

A faster version of Veo 3.1 optimized for lower latency while maintaining high-quality video and audio output.

• Third-party
• Zero data retention

tts-1.5-max

Highest-quality text-to-speech with under 200ms latency, emotion control, and 15-language support.

• Third-party
• Zero data retention

tts-1.5-mini

Ultra-fast, cost-efficient text-to-speech with approximately 120ms latency and 15-language support.

• Third-party
• Zero data retention

tts-2

Inworld's most powerful and expressive text-to-speech model. Builds on TTS 1.5 with rich expressive speech, real-time latency, natural language steering (e.g. [whisper], [say excitedly]), and stronger multilingual support across 15 production languages plus 90+ experimental languages.

• Third-party
• Zero data retention

More than 2x the size of Medium, with softer post-training. Outputs are rawer, more textured, and more flexible — at its best, Large produces results Medium can't match. Strongest on photorealism, raw looks (motion blur, grain, low dynamic range), and expressive and artistic styles.

• Third-party

Smaller, faster, more cost-efficient. Extensive post-training makes outputs especially stable and consistent across generations. Strongest on illustration, anime, painting, and other expressive or artistic styles.

• Third-party

The fastest Krea 2 model, built for low-cost iteration on expressive illustrations, style-driven concepts, and rapid visual exploration. Keeps the Krea 2 style system and expressive visual range but uses a distilled sampling schedule so you can move through ideas much faster. Especially useful for expressive illustration, graphic styles, typography experiments, and quick campaign or concept directions.

• Third-party

hailuo-2.3

A high-fidelity video generation model optimized for realistic human motion, cinematic VFX, expressive characters, and strong prompt and style adherence across text-to-video and image-to-video workflows.

• Third-party
• Zero data retention

hailuo-2.3-fast

A lower-latency version of Hailuo 2.3 that preserves core motion quality, visual consistency, and stylization while enabling faster iteration.

• Third-party
• Zero data retention

m2.7

MiniMax's M2.7 language model with multilingual capabilities.

• Third-party
• Zero data retention

m3

MiniMax's M3 language model with frontier coding and agentic capabilities, a 1M token context window, and multilingual support.

• Third-party
• Zero data retention

music-2.6

MiniMax's music generation model that creates full-length songs with vocals from text prompts and lyrics, or instrumental tracks. Supports BPM/key control and auto-generated lyrics.

• Third-party
• Zero data retention

speech-2.8-hd

MiniMax Speech 2.8 HD focuses on studio-grade audio generation with emotion control, multilingual support (40+ languages), and voice cloning.

• Third-party
• Zero data retention

speech-2.8-turbo

MiniMax Speech 2.8 Turbo turns text into natural, expressive speech with voice cloning, emotion control, and 40+ language support at faster speeds.

• Third-party
• Zero data retention

kimi-k3

Kimi K3 is Moonshot's flagship 2.8 trillion-parameter model, built on Kimi Delta Attention (a hybrid linear attention mechanism) with Attention Residuals. It offers native visual understanding, always-on reasoning, and a 1M-token context window for long-horizon coding, knowledge work, and deep reasoning tasks.

• Third-party

gpt-4.1

OpenAI's flagship GPT model for complex tasks with a million-token context window.

• Third-party
• Zero data retention

gpt-4.1-mini

Fast, affordable version of GPT-4.1 with a million-token context window.

• Third-party
• Zero data retention

gpt-4.1-nano

GPT-4.1 Nano is OpenAI’s smallest and cheapest GPT-4.1 variant, optimized for high-throughput, low-latency tasks.

• Third-party
• Zero data retention

gpt-4o

GPT-4o is OpenAI’s multimodal flagship, accepting text and images and responding quickly across a wide range of tasks.

• Third-party
• Zero data retention

gpt-4o-mini

GPT-4o Mini is the lightweight, low-cost variant of GPT-4o, well suited to high-volume tasks with multimodal inputs.

• Third-party
• Zero data retention

gpt-4o-transcribe

A speech-to-text model that uses GPT-4o to transcribe audio with improved word error rate and better language recognition compared to original Whisper models.

• Third-party
• Zero data retention

gpt-5

OpenAI's model excelling at coding, writing, and reasoning.

• Third-party
• Zero data retention

gpt-5-mini

GPT-5 Mini is the lightweight, low-cost variant of GPT-5, well suited to high-volume coding and reasoning tasks.

• Third-party
• Zero data retention

gpt-5-nano

GPT-5 Nano is OpenAI’s smallest GPT-5 variant, optimized for low latency and cheap, high-throughput tasks.

• Third-party
• Zero data retention

gpt-5.1

GPT-5.1 is OpenAI’s incremental improvement over GPT-5, with stronger coding, reasoning, and writing.

• Third-party
• Zero data retention

gpt-5.4

GPT-5.4 is OpenAI's flagship model with strong coding, reasoning, and multimodal capabilities.

• Third-party
• Zero data retention

gpt-5.4-mini

GPT-5.4 mini is a smaller, faster, and more cost-efficient version of GPT-5.4 for lightweight tasks.

• Third-party
• Zero data retention

gpt-5.4-nano

GPT-5.4 nano is OpenAI's smallest and fastest model, optimized for edge and low-latency use cases.

• Third-party
• Zero data retention

gpt-5.4-pro

GPT-5.4 pro uses OpenAI's Responses API with built-in tools, improved reasoning, and stateful context management.

• Third-party
• Zero data retention

gpt-5.5

GPT-5.5 is OpenAI's flagship model with strong coding, reasoning, and multimodal capabilities.

• Third-party
• Zero data retention

gpt-5.5-pro

GPT-5.5 pro uses OpenAI's Responses API with built-in tools, improved reasoning, and stateful context management.

• Third-party
• Zero data retention

gpt-5.6-luna

GPT-5.6 Luna is an OpenAI GPT-5.6 model optimized for cost-sensitive workloads, using the Responses API for efficient text generation.

• Third-party

gpt-5.6-sol

GPT-5.6 Sol is OpenAI's frontier GPT-5.6 model for complex professional work, using the Responses API for reasoning and stateful context management.

• Third-party

gpt-5.6-terra

GPT-5.6 Terra is an OpenAI GPT-5.6 model that balances intelligence and cost, using the Responses API for reasoning and stateful context management.

• Third-party

gpt-image-1.5

OpenAI's image generation model that creates and edits images from text prompts, supporting multiple quality levels and output sizes.

• Third-party
• Zero data retention

gpt-image-2

OpenAI's next-generation image model that creates and edits images from text prompts, with support for multiple quality levels, sizes, and output formats. Note: transparent backgrounds are not supported — use openai/gpt-image-1.5 for transparent PNGs.

• Third-party
• Zero data retention

o3

o3 is OpenAI’s general-purpose reasoning model, balancing strong analytical performance with reasonable latency and cost.

• Third-party
• Zero data retention

o3-mini

o3-mini is the lightweight, low-cost reasoning variant of o3, well suited to quick analytical tasks at scale.

• Third-party
• Zero data retention

o4-mini

OpenAI's fast, lightweight reasoning model optimized for multi-step problem solving at lower cost.

• Third-party
• Zero data retention

tts-1

OpenAI's text-to-speech model optimized for real-time use with low latency.

• Third-party
• Zero data retention

tts-1-hd

OpenAI's high-definition text-to-speech model producing higher quality audio output.

• Third-party
• Zero data retention

v5.6

Pixverse v5.6 is a video generation model supporting text-to-video and image-to-video with audio generation, customizable aspect ratios, and up to 1080p output.

• Third-party
• Zero data retention

v6

Pixverse v6 is the latest Pixverse video model with support for up to 15-second videos, customizable duration from 1 to 15 seconds, and audio generation.

• Third-party
• Zero data retention

p-image

Pruna's P-Image is an ultra-fast text-to-image model with automatic prompt enhancement and 2-stage refinement, combining exceptional speed with high-quality output and flexible aspect ratios.

• Third-party

p-image-edit

Pruna's P-Image-Edit edits and composes 1-5 reference images with text instructions. It supports complex compositions, style transfers, and targeted edits with flexible output aspect ratios.

• Third-party

p-image-try-on

Pruna's P-Image Try-On virtually fits one or more garments onto a person's photo. Provide a photo of a person plus garment reference images and the model realistically dresses the person in the provided garments.

• Third-party

p-image-upscale

Pruna's P-Image-Upscale increases image resolution using AI, targeting 1-128 megapixels with optional detail and realism enhancement for sharper, cleaner results.

• Third-party

p-video

Pruna's P-Video is a premium video generation model supporting text-to-video, image-to-video, and audio-conditioned generation up to 1080p at 24 or 48 fps, with configurable duration up to 20 seconds.

• Third-party

p-video-animate

Pruna's P-Video-Animate takes a source video and a subject reference image, then animates the referenced subject using the motion and audio from the source video.

• Third-party

p-video-avatar

Pruna's P-Video-Avatar generates talking-head videos from a single portrait image driven by a text script or audio file, with multiple voices, languages, and output resolutions.

• Third-party

p-video-replace

Pruna's P-Video-Replace takes a source video and one or more identity reference images, then places the referenced person or people into the video while preserving the source motion and audio.

• Third-party

recraftv3

Recraft V3 is the previous-generation text-to-image model from Recraft, well-suited to design-quality compositions, brand-aware imagery, and accurate text rendering.

• Third-party
• Zero data retention

recraftv4

Recraft V4 generates art-directed images with strong composition, accurate text rendering, and design taste built in. Fast and cost-efficient at standard resolution.

• Third-party
• Zero data retention

recraftv4-1

Recraft V4.1 generates art-directed images tuned for high aesthetics, with strong composition, accurate text rendering, and refined design taste. Fast and cost-efficient at standard resolution.

• Third-party
• Zero data retention

recraftv4-1-pro

Recraft V4.1 Pro generates high-resolution, art-directed images at 2048px+ tuned for high aesthetics, with strong composition, text rendering, and refined design taste. Built for print and production work.

• Third-party
• Zero data retention

recraftv4-1-pro-vector

Generate detailed, high-resolution SVG vector graphics from text prompts with high aesthetic quality, fine geometry, scalable to any size for print and design work.

• Third-party
• Zero data retention

recraftv4-1-utility

Recraft V4.1 Utility is a general-purpose text-to-image model balancing quality and flexibility for a wide range of everyday use cases at standard resolution.

• Third-party
• Zero data retention

recraftv4-1-utility-pro

Recraft V4.1 Utility Pro is a general-purpose text-to-image model producing high-resolution 2048px+ output for a wide range of production and print use cases.

• Third-party
• Zero data retention

recraftv4-1-utility-pro-vector

Generate detailed, high-resolution SVG vector graphics from text prompts with a general-purpose model, scalable to any size for print and large-scale design work.

• Third-party
• Zero data retention

recraftv4-1-utility-vector

Generate production-ready SVG vector graphics from text prompts with a general-purpose model suited for a wide range of design and illustration tasks.

• Third-party
• Zero data retention

recraftv4-1-vector

Generate production-ready SVG vector graphics from text prompts with high aesthetic quality, clean geometry, structured layers, and editable paths.

• Third-party
• Zero data retention

recraftv4-pro

Recraft V4 Pro generates high-resolution, art-directed images at 2048px+ with strong composition, text rendering, and design taste. Built for print and production work.

• Third-party
• Zero data retention

recraftv4-pro-vector

Generate detailed, production-ready SVG vector graphics from text prompts with fine geometry, scalable to any size for print and design work.

• Third-party
• Zero data retention

recraftv4-vector

Generate production-ready SVG vector graphics from text prompts with clean geometry, structured layers, and editable paths.

• Third-party
• Zero data retention

aleph-2

RunwayML's video editing model. Edit one frame to update your whole video, make changes across multiple shots, and work with up to 30 seconds of video. Supports keyframe-guided editing for precise control over specific moments in the clip.

• Third-party

gen-4.5

RunwayML's video generation model supporting both text-to-video and image-to-video with customizable duration, aspect ratio, and content moderation controls.

• Third-party

Inkling is Thinking Machines' open-weights hybrid reasoning model, built on a mixture-of-experts architecture. It reasons by default, exposing its chain-of-thought as leading thinking content blocks, with reasoning effort tunable via a Tinker-specific output_config.effort parameter. Available through Tinker's beta Anthropic Messages-compatible endpoint alongside tool use, streaming, and multi-turn conversations. Currently intended for low-traffic testing and internal use rather than high-throughput production deployments; prompt caching, citations, and audio input are not supported through this endpoint.

• Third-party

The 256K-context variant of Inkling, Thinking Machines' open-weights hybrid reasoning MoE model. Same hybrid reasoning, tool-use, and streaming support as the base model, with an extended context window for longer conversations and documents. Currently intended for low-traffic testing and internal use rather than high-throughput production deployments.

• Third-party

q3-pro

Vidu Q3 Pro is a high-quality video generation model supporting text-to-video, image-to-video, and start/end-frame-to-video workflows with audio and up to 16-second clips.

• Third-party
• Zero data retention

q3-turbo

Vidu Q3 Turbo is a faster version of Vidu Q3 optimized for lower latency video generation while maintaining audio support and up to 16-second clips.

• Third-party
• Zero data retention

grok-4.20-0309-non-reasoning

xAI's Grok 4.20 non-reasoning model. Skips the thinking trace for fast, single-pass responses while keeping the same training as the reasoning variant.

• Third-party

grok-4.20-0309-reasoning

xAI's Grok 4.20 reasoning model. Uses extended thinking to work through complex problems, returning a reasoning trace alongside the final answer.

• Third-party

grok-4.20-multi-agent-0309

xAI's Grok 4.20 multi-agent model with a 2M-token context window. Multiple agents collaborate in parallel to perform deep research tasks, with function calling, structured outputs, and reasoning capabilities.

• Third-party

grok-4.3

xAI's Grok 4.3 model with a 1M-token context window and strong agentic tool calling with minimal hallucinations. Accepts text and image inputs, and supports function calling, structured outputs, and configurable reasoning effort (none, low, medium, high).

• Third-party

grok-4.5

xAI's Grok 4.5, a frontier model built for coding, agentic tasks, and knowledge work. Accepts text and image inputs, and supports function calling, structured outputs, and configurable reasoning effort (low, medium, high).

• Third-party

grok-imagine-image

xAI's Grok Imagine image model. Generates and edits images from text and reference-image inputs with configurable aspect ratio and resolution.

• Third-party

grok-imagine-image-quality

xAI's higher-fidelity text-to-image model optimized for sharper details, more accurate compositions, and stronger text rendering. Supports image editing via reference images and masks. Trades speed for quality compared to grok-imagine-image. Default output at 2k resolution.

• Third-party

grok-imagine-video

xAI's video generation model. Generates, edits, and extends videos from text and image inputs with native synchronized audio including dialogue, sound effects, and music. Supports multiple creative modes (normal, fun, custom).

• Third-party

grok-imagine-video-1.5-preview

xAI's next-generation video generation model. Generates, edits, and extends videos from text and image inputs. Supports multiple aspect ratios and resolutions with improved quality over the previous generation.

• Third-party

grok-stt

xAI's Grok speech-to-text model. Transcribes audio files into text across 25 languages with word-level timestamps, multichannel transcription, speaker diarization, and key-term biasing.

• Third-party

grok-tts

xAI's Grok text-to-speech model. Generates high-fidelity spoken audio in 5 expressive voices (eve, ara, rex, sal, leo) with 20+ supported languages. Supports inline speech tags for laughter, whispers, and pauses.

• Third-party

grok-voice

xAI's real-time voice conversation model with low-latency audio input and output streaming.

• Third-party

aura-1

Aura is a context-aware text-to-speech (TTS) model that applies natural pacing, expressiveness, and fillers based on the context of the provided text. The quality of your text input directly impacts the naturalness of the audio output.

• Cloudflare-hosted
• Batch
• Partner
• Real-time

aura-2-en

Aura-2 is a context-aware text-to-speech (TTS) model that applies natural pacing, expressiveness, and fillers based on the context of the provided text. The quality of your text input directly impacts the naturalness of the audio output.

• Cloudflare-hosted
• Batch
• Partner
• Real-time

aura-2-es

Aura-2 is a context-aware text-to-speech (TTS) model that applies natural pacing, expressiveness, and fillers based on the context of the provided text. The quality of your text input directly impacts the naturalness of the audio output.

• Cloudflare-hosted
• Batch
• Partner
• Real-time

bart-large-cnn

Beta

BART is a transformer encoder-encoder (seq2seq) model with a bidirectional (BERT-like) encoder and an autoregressive (GPT-like) decoder. You can use this model for text summarization.

• Cloudflare-hosted
• Deprecated

bge-base-en-v1.5

BAAI general embedding (Base) model that transforms any given text into a 768-dimensional vector

• Cloudflare-hosted
• Batch

bge-large-en-v1.5

BAAI general embedding (Large) model that transforms any given text into a 1024-dimensional vector

• Cloudflare-hosted
• Batch

bge-m3

Multi-Functionality, Multi-Linguality, and Multi-Granularity embeddings model.

• Cloudflare-hosted

bge-reranker-base

Different from embedding model, reranker uses question and document as input and directly output similarity instead of embedding. You can get a relevance score by inputting query and passage to the reranker. And the score can be mapped to a float value in [0,1] by sigmoid function.

• Cloudflare-hosted

bge-small-en-v1.5

BAAI general embedding (Small) model that transforms any given text into a 384-dimensional vector

• Cloudflare-hosted
• Batch

deepseek-r1-distill-qwen-32b

DeepSeek-R1-Distill-Qwen-32B is a model distilled from DeepSeek-R1 based on Qwen2.5. It outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.

• Cloudflare-hosted
• Reasoning

detr-resnet-50

Beta

DEtection TRansformer (DETR) model trained end-to-end on COCO 2017 object detection (118k annotated images).

• Cloudflare-hosted

distilbert-sst-2-int8

Distilled BERT model that was finetuned on SST-2 for sentiment classification

• Cloudflare-hosted

Stable Diffusion model that has been fine-tuned to be better at photorealism without sacrificing range.

• Cloudflare-hosted

embeddinggemma-300m

EmbeddingGemma is a 300M parameter, state-of-the-art for its size, open embedding model from Google, built from Gemma 3 (with T5Gemma initialization) and the same research and technology used to create Gemini models. EmbeddingGemma produces vector representations of text, making it well-suited for search and retrieval tasks, including classification, clustering, and semantic similarity search. This model was trained with data in 100+ spoken languages.

• Cloudflare-hosted

flux

Flux is the first conversational speech recognition model built specifically for voice agents.

• Cloudflare-hosted
• Partner
• Real-time

flux-1-schnell

FLUX.1 [schnell] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions.

• Cloudflare-hosted

flux-2-dev

FLUX.2 [dev] is an image model from Black Forest Labs where you can generate highly realistic and detailed images, with multi-reference support.

• Cloudflare-hosted
• Partner

flux-2-klein-4b

FLUX.2 [klein] is an ultra-fast, distilled image model. It unifies image generation and editing in a single model, delivering state-of-the-art quality enabling interactive workflows, real-time previews, and latency-critical applications.

• Cloudflare-hosted
• Partner

flux-2-klein-9b

FLUX.2 [klein] 9B is an ultra-fast, distilled image model with enhanced quality. It unifies image generation and editing in a single model, delivering state-of-the-art quality enabling interactive workflows, real-time previews, and latency-critical applications.

• Cloudflare-hosted
• Partner

gemma-2b-it-lora

Beta

This is a Gemma-2B base model that Cloudflare dedicates for inference with LoRA adapters. Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models.

• Cloudflare-hosted
• LoRA

gemma-3-12b-it

Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Gemma 3 models are multimodal, handling text and image input and generating text output, with a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions.

• Cloudflare-hosted
• LoRA
• Deprecated

gemma-4-26b-a4b-it

Gemma 4 is Google's most intelligent family of open models, built from Gemini 3 research to maximize intelligence-per-parameter.

• Cloudflare-hosted
• Function calling
• Reasoning
• Vision

gemma-7b-it

Beta

Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. They are text-to-text, decoder-only large language models, available in English, with open weights, pre-trained variants, and instruction-tuned variants.

• Cloudflare-hosted
• LoRA
• Deprecated

gemma-7b-it-lora

Beta

This is a Gemma-7B base model that Cloudflare dedicates for inference with LoRA adapters. Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models.

• Cloudflare-hosted
• LoRA

SEA-LION stands for Southeast Asian Languages In One Network, which is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region.

• Cloudflare-hosted

glm-4.7-flash

GLM-4.7-Flash is a fast and efficient multilingual text generation model with a 131,072 token context window. Optimized for dialogue, instruction-following, and multi-turn tool calling across 100+ languages.

• Cloudflare-hosted
• Function calling
• Reasoning

glm-5.2

Z.ai's flagship agentic coding model

• Cloudflare-hosted
• Function calling
• Reasoning

gpt-oss-120b

OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases – gpt-oss-120b is for production, general purpose, high reasoning use-cases.

• Cloudflare-hosted
• Function calling
• Reasoning

gpt-oss-20b

OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases – gpt-oss-20b is for lower latency, and local or specialized use-cases.

• Cloudflare-hosted
• Function calling
• Reasoning

granite-4.0-h-micro

Granite 4.0 instruct models deliver strong performance across benchmarks, achieving industry-leading results in key agentic tasks like instruction following and function calling. These efficiencies make the models well-suited for a wide range of use cases like retrieval-augmented generation (RAG), multi-agent workflows, and edge deployments.

• Cloudflare-hosted
• Function calling

Hermes 2 Pro on Mistral 7B is the new flagship 7B Hermes! Hermes 2 Pro is an upgraded, retrained version of Nous Hermes 2, consisting of an updated and cleaned version of the OpenHermes 2.5 Dataset, as well as a newly introduced Function Calling and JSON Mode dataset developed in-house.

• Cloudflare-hosted
• Function calling
• Deprecated

IndicTrans2 is the first open-source transformer-based multilingual NMT model that supports high-quality translations across all the 22 scheduled Indic languages

• Cloudflare-hosted

kimi-k2.5

Kimi K2.5 is a frontier-scale open-source model with a 256k context window, multi-turn tool calling, vision inputs, and structured outputs for agentic workloads.

• Cloudflare-hosted
• Function calling
• Deprecated
• Reasoning
• Vision

kimi-k2.6

Kimi K2.6 is a frontier-scale open-source 1T parameter model with a 262.1k context window, multi-turn tool calling, vision inputs, and structured outputs for agentic workloads.

• Cloudflare-hosted
• Function calling
• Reasoning
• Vision

kimi-k2.7-code

Kimi K2.7 is a frontier-scale open-source 1T parameter model with a 262.1k context window, multi-turn tool calling, vision inputs, and structured outputs for agentic workloads.

• Cloudflare-hosted
• Function calling
• Reasoning
• Vision

llama-2-7b-chat-fp16

Full precision (fp16) generative text model with 7 billion parameters from Meta

• Cloudflare-hosted
• Deprecated

llama-2-7b-chat-hf-lora

Beta

This is a Llama2 base model that Cloudflare dedicated for inference with LoRA adapters. Llama 2 is a collection of pretrained and fine-tuned generative text models ranging in scale from 7 billion to 70 billion parameters. This is the repository for the 7B fine-tuned model, optimized for dialogue use cases and converted for the Hugging Face Transformers format.

• Cloudflare-hosted
• LoRA

llama-2-7b-chat-int8

Quantized (int8) generative text model with 7 billion parameters from Meta

• Cloudflare-hosted
• Deprecated

llama-3-8b-instruct

Generation over generation, Meta Llama 3 demonstrates state-of-the-art performance on a wide range of industry benchmarks and offers new capabilities, including improved reasoning.

• Cloudflare-hosted
• Deprecated

llama-3-8b-instruct-awq

Quantized (int4) generative text model with 8 billion parameters from Meta.

• Cloudflare-hosted
• Deprecated

llama-3.1-70b-instruct

The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models. The Llama 3.1 instruction tuned text only models are optimized for multilingual dialogue use cases and outperform many of the available open source and closed chat models on common industry benchmarks.

• Cloudflare-hosted
• Deprecated

llama-3.1-8b-instruct

The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models. The Llama 3.1 instruction tuned text only models are optimized for multilingual dialogue use cases and outperform many of the available open source and closed chat models on common industry benchmarks.

• Cloudflare-hosted
• Deprecated

llama-3.1-8b-instruct-awq

Quantized (int4) generative text model with 8 billion parameters from Meta.

• Cloudflare-hosted
• Deprecated

llama-3.1-8b-instruct-fast

[Fast version] The Meta Llama 3.1 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction tuned generative models. The Llama 3.1 instruction tuned text only models are optimized for multilingual dialogue use cases and outperform many of the available open source and closed chat models on common industry benchmarks.

• Cloudflare-hosted

llama-3.1-8b-instruct-fp8

Llama 3.1 8B quantized to FP8 precision

• Cloudflare-hosted

llama-3.2-11b-vision-instruct

The Llama 3.2-Vision instruction-tuned models are optimized for visual recognition, image reasoning, captioning, and answering general questions about an image.

• Cloudflare-hosted
• LoRA
• Vision

llama-3.2-1b-instruct

The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks.

• Cloudflare-hosted

llama-3.2-3b-instruct

The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks.

• Cloudflare-hosted

llama-3.3-70b-instruct-fp8-fast

Llama 3.3 70B quantized to fp8 precision, optimized to be faster.

• Cloudflare-hosted
• Batch
• Function calling

llama-4-scout-17b-16e-instruct

Meta's Llama 4 Scout is a 17 billion parameter model with 16 experts that is natively multimodal. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.

• Cloudflare-hosted
• Batch
• Function calling
• Vision

llama-guard-3-8b

Llama Guard 3 is a Llama-3.1-8B pretrained model, fine-tuned for content safety classification. Similar to previous versions, it can be used to classify content in both LLM inputs (prompt classification) and in LLM responses (response classification). It acts as an LLM – it generates text in its output that indicates whether a given prompt or response is safe or unsafe, and if unsafe, it also lists the content categories violated.

• Cloudflare-hosted
• LoRA

LLaVA is an open-source chatbot trained by fine-tuning LLaMA/Vicuna on GPT-generated multimodal instruction-following data. It is an auto-regressive language model, based on the transformer architecture.

• Cloudflare-hosted

lucid-origin

Lucid Origin from Leonardo.AI is their most adaptable and prompt-responsive model to date. Whether you're generating images with sharp graphic design, stunning full-HD renders, or highly specific creative direction, it adheres closely to your prompts, renders text with accuracy, and supports a wide array of visual styles and aesthetics – from stylized concept art to crisp product mockups.

• Cloudflare-hosted
• Partner

m2m100-1.2b

Multilingual encoder-decoder (seq-to-seq) model trained for Many-to-Many multilingual translation

• Cloudflare-hosted
• Batch

melotts

MeloTTS is a high-quality multi-lingual text-to-speech library by MyShell.ai.

• Cloudflare-hosted

meta-llama-3-8b-instruct

Generation over generation, Meta Llama 3 demonstrates state-of-the-art performance on a wide range of industry benchmarks and offers new capabilities, including improved reasoning.

• Cloudflare-hosted
• Deprecated

mistral-7b-instruct-v0.1

Instruct fine-tuned version of the Mistral-7b generative text model with 7 billion parameters

• Cloudflare-hosted
• LoRA
• Deprecated

mistral-7b-instruct-v0.2

Beta

The Mistral-7B-Instruct-v0.2 Large Language Model (LLM) is an instruct fine-tuned version of the Mistral-7B-v0.2. Mistral-7B-v0.2 has the following changes compared to Mistral-7B-v0.1: 32k context window (vs 8k context in v0.1), rope-theta = 1e6, and no Sliding-Window Attention.

• Cloudflare-hosted
• LoRA
• Deprecated

mistral-7b-instruct-v0.2-lora

Beta

The Mistral-7B-Instruct-v0.2 Large Language Model (LLM) is an instruct fine-tuned version of the Mistral-7B-v0.2.

• Cloudflare-hosted
• LoRA

mistral-small-3.1-24b-instruct

Building upon Mistral Small 3 (2501), Mistral Small 3.1 (2503) adds state-of-the-art vision understanding and enhances long context capabilities up to 128k tokens without compromising text performance. With 24 billion parameters, this model achieves top-tier capabilities in both text and vision tasks.

• Cloudflare-hosted
• Function calling

Moondream 3 is a fast, efficient 9B mixture-of-experts vision language model (2B active parameters) that delivers frontier-level visual reasoning for tasks like object detection, pointing, OCR, and structured output.

• Cloudflare-hosted
• Vision

nemotron-3-120b-a12b

NVIDIA Nemotron 3 Super is a hybrid MoE model with leading accuracy for multi-agent applications and specialized agentic AI systems.

• Cloudflare-hosted
• Function calling
• Reasoning

nova-3

Transcribe audio using Deepgram’s speech-to-text model

• Cloudflare-hosted
• Batch
• Partner
• Real-time

phi-2

Beta

Phi-2 is a Transformer-based model with a next-word prediction objective, trained on 1.4T tokens from multiple passes on a mixture of Synthetic and Web datasets for NLP and coding.

• Cloudflare-hosted
• Deprecated

phoenix-1.0

Phoenix 1.0 is a model by Leonardo.Ai that generates images with exceptional prompt adherence and coherent text.

• Cloudflare-hosted
• Partner

PLaMo-Embedding-1B is a Japanese text embedding model developed by Preferred Networks, Inc. It can convert Japanese text input into numerical vectors and can be used for a wide range of applications, including information retrieval, text classification, and clustering.

• Cloudflare-hosted

qwen2.5-coder-32b-instruct

Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). As of now, Qwen2.5-Coder has covered six mainstream model sizes, 0.5, 1.5, 3, 7, 14, 32 billion parameters, to meet the needs of different developers. Qwen2.5-Coder brings the following improvements upon CodeQwen1.5:

• Cloudflare-hosted
• LoRA

qwen3-30b-a3b-fp8

Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support.

• Cloudflare-hosted
• Batch
• Function calling
• Reasoning

qwen3-embedding-0.6b

The Qwen3 Embedding model series is the latest proprietary model of the Qwen family, specifically designed for text embedding and ranking tasks.

• Cloudflare-hosted

qwq-32b

QwQ is the reasoning model of the Qwen series. Compared with conventional instruction-tuned models, QwQ, which is capable of thinking and reasoning, can achieve significantly enhanced performance in downstream tasks, especially hard problems. QwQ-32B is the medium-sized reasoning model, which is capable of achieving competitive performance against state-of-the-art reasoning models, e.g., DeepSeek-R1, o1-mini.

• Cloudflare-hosted
• LoRA
• Reasoning

resnet-50

50 layers deep image classification CNN trained on more than 1M images from ImageNet

• Cloudflare-hosted

smart-turn-v2

An open source, community-driven, native audio turn detection model in 2nd version

• Cloudflare-hosted
• Batch
• Real-time

sqlcoder-7b-2

Beta

This model is intended to be used by non-technical users to understand data inside their SQL databases.

• Cloudflare-hosted
• Deprecated

stable-diffusion-v1-5-img2img

Beta

Stable Diffusion is a latent text-to-image diffusion model capable of generating photo-realistic images. Img2img generate a new image from an input image with Stable Diffusion.

• Cloudflare-hosted

stable-diffusion-v1-5-inpainting

Beta

Stable Diffusion Inpainting is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input, with the extra capability of inpainting the pictures by using a mask.

• Cloudflare-hosted

stable-diffusion-xl-base-1.0

Beta

Diffusion-based text-to-image generative model by Stability AI. Generates and modify images based on text prompts.

• Cloudflare-hosted

stable-diffusion-xl-lightning

Beta

SDXL-Lightning is a lightning-fast text-to-image generation model. It can generate high-quality 1024px images in a few steps.

• Cloudflare-hosted

uform-gen2-qwen-500m

Beta

UForm-Gen is a small generative vision-language model primarily designed for Image Captioning and Visual Question Answering. The model was pre-trained on the internal image captioning dataset and fine-tuned on public instructions datasets: SVIT, LVIS, VQAs datasets.

• Cloudflare-hosted
• Deprecated

whisper

Whisper is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio and is also a multitasking model that can perform multilingual speech recognition, speech translation, and language identification.

• Cloudflare-hosted

whisper-large-v3-turbo

Whisper is a pre-trained model for automatic speech recognition (ASR) and speech translation.

• Cloudflare-hosted
• Batch

whisper-tiny-en

Beta

Whisper is a pre-trained model for automatic speech recognition (ASR) and speech translation. Trained on 680k hours of labelled data, Whisper models demonstrate a strong ability to generalize to many datasets and domains without the need for fine-tuning. This is the English-only version of the Whisper Tiny model which was trained on the task of speech recognition.

• Cloudflare-hosted