Compare AI models by price, context and where your data goes

Preview build of the Jev classification model: typed yes/no, choice, score and box questions over text, JSON or image state. Images and box questions are beta. 64k context; state plus longest question limited to 32k tokens.

Meta's frontier model for coding, long-running agents, and computer use, with active memory and a one-million-token context window.

Microsoft's decision model returns calibrated probabilities for yes/no, multiple-choice, and rating questions over supplied text.

Cloudflare clef-omni returns typed decisions and probabilities for questions over application state.

Cloudflare clef returns typed decisions and probabilities for questions over application state.

Qwen-Image-2.1's accelerated 7B visual generation checkpoint for text-to-image generation and image editing in 8 denoising steps.

Qwen-Image-2.1 Pro supports text-to-image generation, transparent images, and editing with up to 10 reference images.

Cloudflare clef-flash returns typed decisions and probabilities for questions over application state.

Liquid AI decision model returning calibrated Noul, Choice, and Score answers with zero generated tokens.

Claude Haiku 5.5

Claude Mythos 5.1

Qwen3.5-2B System One answers typed classification questions over application state through the System One API.

laya answers typed classification questions over application state through the System One API.

Nano Banana 2.1 image generation and conversational editing model with improved visual quality, subject consistency, and text rendering.

Multimodal model with 49B active parameters, 1.05T total parameters, and adjustable reasoning effort.

Qwen 3.8 27B is a multimodal reasoning model with tool use, structured outputs, and long-context support.

One-pass decision model: typed yes/no, choice and score questions over text or JSON state, with calibrated probabilities. State limited to 32k tokens; up to 255 options per choice.

Black Forest Labs image generation and editing model with native 2K and 4K output, precise local edits, and up to 10 reference images.

Decision model that answers yes/no, multiple-choice, and rubric-scoring questions about text, JSON, and images with probabilities.

Standard Sonar model with 128k context window, optimized for general-purpose chat completions with web search capabilities.

Low-latency multimodal embedding model for text, images, and mixed text/image PDFs in over 100 languages, sharing an embedding space with Embed 5 Pro.

Multimodal embedding model optimized for retrieval quality across text, images, and mixed text/image PDFs in over 100 languages.

Kimi K2.8 Preview supports reasoning, coding, and tool calling.

GPT-6.1 Sol is OpenAI's reasoning model for complex coding, computer use, and professional work.

Structured classification, yes/no decisions, and scoring with probability distributions and confidence through the System One protocol.

Claude Sonnet 5.5

ByteDance Dola Seedream 5.0 Flash supports text-to-image generation and image editing.

ByteDance Seedream 5.0 Lite supports text-to-image generation and image editing.

Expressive text-to-speech model with audio tags, natural multi-speaker dialogue, and support for more than 90 languages.

Low-latency expressive text-to-speech model with audio tags and support for more than 90 languages, designed for conversational agents and interactive voice experiences.

Holo4 27B is a dense vision-language model for computer-use agents across graphical interfaces, code, and tools.

Holo4 35B A3B is a vision-language Mixture-of-Experts model with 35B total and 3B active parameters for computer-use agents across graphical interfaces, code, and tools.

Tencent Hy-MT2-Lite is a translation model.

Tencent Hy-MT2-Pro is a translation model.

Tencent Hy3 supports reasoning, coding, and agent tasks.

Tencent Hy4 Preview supports reasoning, coding, and agent tasks.

GLM-4-32B-0414-128K is a 32B-parameter text model for tool use, structured output, and code generation.

Claude Opus 5.5

GPT-6 Luna is optimized for focused, high-volume tasks.

GPT-6 Sol is built for complex coding and agentic workflows.

GLM-5.3-Flash is Z.ai's native multimodal GLM-5 model for efficient coding, agentic workflows, visual understanding, and professional document tasks.

Jev classification model: typed yes/no, choice and score questions over text, JSON or image state. Images are beta. 64k context; state plus longest question limited to 32k tokens.

Page-to-Markdown document parser for images and PDFs.

DeepSeek V4.1 Flash is a 552B-parameter MoE model with a Causal-Encoder-Decoder architecture, activating 8B parameters for input and 16B for output, native visual understanding, and a 1M-token context window.

Fast image generation and editing from text and image inputs.

Image generation and editing from text and image inputs, with a focus on editing precision.

GPT-6 Astra is OpenAI's flagship reasoning model for complex professional and agentic work.

Gemini 3.8 Flash is our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows—all with the speed and cost efficiency of Flash.

Meta's multimodal reasoning model for coding and long-running agentic workflows, with a one-million-token context window, tool calling, and structured outputs.

Claude Fable 5.1

Speech-to-text model for transcribing supported WAV recordings with optional language bias.

Nano Banana 2, the stable Gemini 3.1 Flash Image generation and conversational editing model.

Qwen3.8 Flash-Next is an ultra-efficient multimodal model with elite coding from 6B active parameters.

A multimodal document parsing model that extracts text, tables, lists, forms, images, captions, and layout into Markdown.

Qwen3 235B A22B Instruct served by Scaleway Generative APIs.

Mistral Small 3.2 24B Instruct served by Scaleway Generative APIs.

Qwen3 Embedding 4B multilingual embedding model for retrieval, classification, and semantic similarity.

Qwen3 Embedding 8B served by Scaleway Generative APIs.

Qwen3 Reranker 4B multilingual reranking model for scoring documents against a query.

Qwen3.5 397B-A17B

BGE Multilingual Gemma2 embedding model served by Scaleway Generative APIs.

Speech-to-text model for precise transcription of prerecorded audio with speaker attribution and word-level timestamps.

Low-latency speech-to-text model for continuous realtime transcription.

GreenPT-branded next-generation chat model with long-context multimodal reasoning. Recommended default for chat completions.

Speech-to-text model for pre-recorded and live transcription.

Advanced speech-to-text model with multilingual transcription options.

Llama 3.3 70B Instruct served by Scaleway Generative APIs.

An agentic image model for generation and editing with accurate text, complex compositions, reference images, and quality preservation across multi-turn edits.

Muse Spark 1.1 is Meta's multimodal reasoning model with a one-million-token context window, tool calling, and structured outputs.

Muse Spark 1.2 is Meta's multimodal reasoning model with a one-million-token context window, tool calling, and structured outputs.

Mistral OCR 4.1 adds block-level confidence scores to paragraph-level bounding box extraction and structural block labels.

Nemotron 3 Super by NVIDIA is a hybrid Mamba-Transformer MoE model with 120B total and 12B active parameters, optimized for agentic reasoning, multi-step task planning, and long-context workloads.

NVIDIA Nemotron 3 Ultra is a 550B hybrid MoE model for multi-agent and complex reasoning workloads.

GLM-5.3 is Z.ai's flagship model for complex software engineering and long-horizon agent tasks, with always-on configurable reasoning and text-only input.

Gemini 3.7 Flash is optimized for agentic workflows, code generation, multimodal reasoning, and token-efficient multi-step tasks.

Meta's multimodal reasoning model with text and image input and native tool calling.

1B embedding model for semantic search, retrieval, and RAG applications.

NVIDIA's multilingual, multimodal moderator for classifying unsafe and toxic prompts, images, and model responses.

NVIDIA's 30B mixture-of-experts model with 3B active parameters, optimized for fast specialized agentic tasks.

DeepSeek V4-Pro is a 1.6T total / 49B active parameter model with 1M context and top-tier agentic reasoning and coding capabilities.

xAI's flagship model for long-running agents, coding, and interactive work.

Expressive text-to-speech model with streaming generation and emotion controls.

Low-latency text-to-speech model with multilingual voice support.

Low-latency text-to-speech model with multilingual voice support.

Expressive text-to-speech model with natural multi-speaker dialogue and support for more than 70 languages.

Compact multilingual text-to-speech model with streaming generation and voice mixing.

Multilingual automatic speech recognition model for fast batch transcription.

Speech-to-text model with built-in speaker diarization for identifying who spoke when.

Streaming speech-to-text model for low-latency realtime transcription.

Qwen3.8 Max is Alibaba's 2.4T-parameter, 95B-active flagship model for coding, professional work, and long-horizon multimodal agents.

Low-latency speech-to-text model for realtime transcription with streaming transcript deltas.

High-accuracy speech-to-text model for completed audio files, streamed file transcripts, and committed turns in Realtime sessions.

Multilingual listwise document reranker with improved domain robustness and inference efficiency

Mistral OCR 4, the latest OCR service powering the Document AI stack. Extracts markdown-structured text with paragraph-level bounding boxes, block classification, and confidence scores across 170 languages from image documents.

Claude Opus 5

Gemini 3.5 Flash-Lite is a low-latency multimodal model optimized for high-throughput agentic workflows and document parsing.

Gemini 3.6 Flash is optimized for code generation, agentic execution, spatial reasoning, and complex multimodal tasks.

Kimi K3 is Moonshot AI’s 2.8-trillion-parameter flagship open-source model for long-horizon coding, reasoning, and end-to-end knowledge work, with native visual understanding and a 1M-token context window.

xAI's frontier model for coding, agentic tasks, and knowledge work.

Gemini 3 Pro Image generation and conversational editing model.

Gemini 3.1 Flash Live Preview for bidirectional realtime sessions.

Gemini 3.1 Flash text-to-speech preview model.

Preview of Qwen 3.6 27B on Groq with text and image input, reasoning, tool use, and JSON object mode.

Qwen3.7 Max

Qwen3.7 Plus

Batch speech recognition model with accurate transcription in 90+ languages, keyterm prompting, entity detection, word-level timestamps, speaker diarization, audio tagging, and language detection.

A lightweight multilingual text embedding model for large-scale retrieval and latency-sensitive workloads.

A multilingual text embedding model for high-quality retrieval and deep semantic understanding.

Fast, reliable model with strong tool calling and instruction following capabilities.

SpaceXAI's intelligent coding model for agentic software, engineering, and workflow tasks.

Qwen3 30B A3B Instruct 2507 is a versatile instruction-tuned MoE model for chat, reasoning, and coding.

Hermes 4 405B is a hybrid reasoning model trained for strong math, coding, and step-by-step reliability.

MiniCPM-V-4.5 is a compact multimodal model for image, video, OCR, and PDF understanding.

Nemotron 3 Nano 30B A3B is a compact MoE model optimized for reasoning, coding, and long-context RAG workflows.

Mistral code model for code generation, chat, and fill-in-the-middle completion.

Codestral Embed, a code-optimized embedding model with 8K context.

Ministral 8B, an efficient edge model with 256K context, vision, and function calling.

Mistral Moderation 2 detects harmful text content and jailbreaking attempts.

Mistral Tiny, a small fast chat model with 131K context.

Voxtral Small, an audio-capable chat model (text $0.10/M, audio $0.004/min input) with 32K context and function calling.

Qwen3-VL 235B-A22B

MiMo v2.5, Xiaomi's multimodal model with 1M context, vision, reasoning and function calling.

GPT-5.6 preview model optimized for fast, cost-efficient, high-volume workloads.

GPT-5.6 flagship preview model for complex professional work, agentic coding, science, and cybersecurity.

GPT-5.6 preview model that balances intelligence and cost for everyday work.

Prerecorded is the Reson8 speech-to-text routing alias for complete audio-file transcription.

Poolside's strongest agentic coding model, a 118B total parameter Mixture-of-Experts model with 8B activated parameters.

Poolside's most capable agentic coding model, a 225B total parameter Mixture-of-Experts model with 23B activated parameters.

Poolside's 33B total parameter Mixture-of-Experts agentic coding model with 3B activated parameters.

Tencent Hy-MT2-Plus on Tencent Cloud, a translation model with strong instruction-following capability.

Claude Sonnet 5

Nano Banana 2 Lite, the low-latency Gemini 3.1 Flash Lite Image generation and editing model.

Preview Gemini model for prompt-driven and reference-guided video generation through the Interactions API.

GLM-5.2 is Z.ai flagship long-horizon text model with 1M context, 128K maximum output tokens, thinking mode, function calling, context caching, and structured JSON output.

Kimi K2.7 Code is Moonshot AI's most capable coding model, with stronger long-context instruction following, higher coding task success rates, native text, image, and video input, enforced thinking mode, and 256k context.

Kimi K2.7 Code is Moonshot AI's most capable coding model, with stronger long-context instruction following, higher coding task success rates, native text, image, and video input, enforced thinking mode, and 256k context.

Claude Fable 5

MiniMax-M3 is a frontier multimodal coding model with a 1M context window, agentic reasoning, and tool use.

Claude Opus 4.8

Command A+ is Cohere's first Mixture of Experts model, combining vision input support, agentic reasoning, and world-class translation capabilities.

Gemini 3.5 Flash provides frontier-level intelligence optimized for real-world agentic tasks at higher speed and lower cost.

Compact multimodal embeddings for text, images, audio, video, and PDF

Multimodal embeddings for text, images, audio, video, and PDF

Gemini 3.1 Flash-Lite is a low-latency, cost-effective multimodal model optimized for high-frequency, lightweight tasks.

Qwen-MT Flash

Qwen-MT Lite

Qwen3.6 Flash

Kimi K2.5 (served via Alibaba Cloud)