OpenRouter

Moonshot's flagship open-weight multimodal reasoning model, post-trained from Kimi K3 for long-horizon coding and agent work, with a 1M-token context.

MiniMax's low-cost coding and agent model with a 1M-token context and tunable thinking depth, on Runware and OpenRouter.

Reflection's open-weight reasoning model for coding and long-horizon agentic work: a sparse mixture of experts with 501B total and 23B active parameters, a 1M-token context and tool calling.

Moondream's vision-language model (9B total, 2B active parameters) for questions about images, captions, reading text in images and visual reasoning.

NVIDIA's open multimodal mixture-of-experts model (120B total, 12B active parameters) for coding, reasoning, tool use and document understanding, with image input and a 1M-token context window.

Meta's open-weight multimodal reasoning model (816B total, 42B active parameters) built for agentic coding, with image input and a 1M-token context window.

Liquid AI's mixture-of-experts reasoning model with 24B total and 2.45B active parameters. It always reasons before answering and is built for agents that call tools.

Google's flagship reasoning model for complex software engineering, deep mathematical and scientific problem solving and autonomous multi-step agents, with a shared 1M-token context.

Meta's frontier reasoning model for complex personal and professional agentic tasks, with image input, tool calling and a 1M-token context window.

Anthropic's fastest model, built for high-volume, latency-sensitive work such as classification, extraction, routing and subagent tasks, with vision, tool use and a 1M-token context window.

Mistral's largest open-weight model, a multimodal mixture-of-experts built for reasoning, coding and agent work, with vision, tool calling and a 512K-token context window.

Anthropic's Sonnet-class model for everyday coding and agentic work, a direct upgrade to Sonnet 5 with vision, tool use and a 1M-token context window.

OpenAI's update to GPT 6 Sol, the cost-efficient high-end GPT-6 tier, for agentic coding and document work, with vision and a 1M-token context window.

GPT 6.1 Sol in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

GPT 6 Astra in OpenAI's pro reasoning mode, which works longer on each answer for the highest quality on hard tasks.

DeepSeek's fast mixture-of-experts model and the first on its Causal Encoder-Decoder architecture, with image input, a 1M-token context and native tool calling.

The high-speed variant of Z.ai's GLM-5.3, with the same capabilities at 1.5 to 2 times the output throughput and a higher price.

Alibaba's flagship Qwen3.8 reasoning model for coding, agentic workflows and document analysis, with image input and a 1M-token context window.

A higher-throughput variant of Alibaba's Qwen3.8 Max, served as a separate tier at a higher price.

Alibaba's omni-modal Qwen3.8 reasoning model built around agentic work, which reads text, images, audio and video.

Meta's multimodal reasoning model for long-running agentic, multi-agent and coding workflows, with image input and a 1M-token context window.

Google's most capable Flash model, with clear gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning, plus vision and a 1M-token context window.

Anthropic's Fable model for the most demanding reasoning and long-horizon agentic work, with always-on adaptive thinking, vision, and a 1M-token context window.

Anthropic's Opus model for long-running agentic coding and knowledge work, with always-on adaptive thinking, vision, and a 1M-token context window.

OpenAI's flagship GPT-6 model for complex reasoning, coding, research, and long-running agent work, with vision and a 1M-token context window.

OpenAI's fast, low-cost GPT-6 tier for high-volume chat, classification, and light agentic work, with vision.

GPT 6 Luna in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

OpenAI's cost-efficient high-end GPT-6 tier for demanding coding and agentic work, with vision and a 1M-token context window.

GPT 6 Sol in OpenAI's pro reasoning mode, which works longer on each answer for higher quality on hard tasks.

xAI's flagship Grok model for coding, agentic tasks, and knowledge work, with vision and a 500K-token context window.

Z.ai's flagship open-weights coding model, with a 1M-token context and stronger long-horizon agentic work than GLM-5.2.

Z.ai's natively multimodal GLM-5.3 tier for efficient coding and long-horizon agent tasks, with image input and tool calling.

Alibaba's fast, low-cost multimodal Qwen3.8 reasoning model for coding assistance, agentic workflows and visual understanding, with a 1M-token context window.

DeepSeek's high-capability V4 model (MIT) with a 1M-token context, dual thinking modes, and stronger agentic performance than V4 Flash.

Moonshot's 2.8T-parameter open-weight mixture-of-experts reasoning model (104B active) with a 1M-token context, built for long-horizon agent work, coding and research.

Google's most efficient multimodal Flash model: an updated reasoning stack with stronger coding and computer-use quality, at the same speed and scale as the rest of the Flash line.

OpenAI's fastest, most affordable GPT-5.6 tier — for high-volume, latency-sensitive, and budget-conscious workloads.

OpenAI's flagship GPT-5.6 tier — leads the family on every benchmark, for frontier coding, long-horizon agentic work, and research.

OpenAI's balanced GPT-5.6 tier — GPT-5.5-level quality at roughly half the cost, for production workloads.

ByteDance's fast Seed 2.1 reasoning model with a 256K context and output window, native tool calling and image input.

Z.ai's flagship long-horizon coding model, with a 1M-token context and deep-thinking reasoning for end-to-end engineering work.

Anthropic's most agentic Sonnet — frontier-level coding and tool use at Sonnet speed and price, with vision.

Anthropic's Claude 5-family model for the most demanding reasoning, coding, and agentic work — with vision and a 128K output budget.

Anthropic's most capable model — frontier reasoning, coding, and agentic tool use with vision.

Balanced Seed 2.0 reasoning model with tool calling and image input at a lower price than Pro.

The smallest, cheapest Seed 2.0 model for fast everyday tasks, with tool calling and image input.

Moonshot's frontier open-weight model (1T parameters) built for long-horizon agentic coding, with a 262K context and native vision.

xAI's frontier Grok model with strong reasoning, coding, and agentic tool use, plus vision and real-time knowledge from X.

OpenAI's frontier model with advanced reasoning and broad multimodal capabilities.

Google's most capable multimodal model with deep reasoning and 1M token context.

Anthropic's fast, cost-efficient model with strong reasoning, coding, and tool use at low latency.

Black Forest Labs' smaller, faster FLUX 3 image model with open weights: text-to-image, image editing and multi-reference generation, with native 4K output.

Alibaba Qwen's hosted Turbo tier of the Qwen Image 2.1 family, an 8-step checkpoint that generates from a prompt and edits or composes from up to 10 reference images at a low flat price. It can rewrite the prompt with an LLM before generating.

Google's updated Nano Banana 2 image model, with sharper graphic composition, steadier subjects across edits and closer prompt following, from text or up to 14 reference images at 1K to 4K.

Black Forest Labs' image model on the FLUX 3 backbone: text-to-image, precise local edits, up to ten references, accurate typography, and native 4K output.

The lower-latency, lower-cost member of Microsoft AI's MAI-Image 2.6 family, built for high-throughput image generation and editing from up to five reference images.

Microsoft AI's precision image model for design-ready visuals, which generates from a prompt and edits or composes from up to five reference images.

Alibaba Qwen's hosted Pro tier of the Qwen Image 2.1 family, a unified model that generates from a prompt and edits or composes from up to 10 reference images. It can rewrite the prompt with an LLM before generating, with an optional thinking mode, and renders from 0.26 to 4.19 megapixels.

ByteDance's fast, low-cost Seedream 5 tier for high-volume image generation and editing, with multi-reference input.

OpenAI's fastest model for high-quality everyday image generation and editing, with strong instruction-following and sharp text rendering from text and image inputs.

OpenAI's most capable image model, built for workflows where editing precision matters most, with strong subject preservation, localized changes and sharp text rendering across iterative edits.

xAI's Grok image model for generation and editing, with quality-aware output control at up to 2K.

The highest-fidelity tier of Microsoft AI's MAI-Image 2.5 family, producing photorealistic and artistic images from a prompt or one reference image.

Alibaba Qwen's unified image generation and editing model, balancing visual quality, layout fidelity, and speed.

Alibaba Qwen's professional-grade image model for dense layouts, small-text rendering, and multilingual visual design.

ByteDance's flagship Seedream 5 model — top-tier aesthetic quality, prompt fidelity, and multi-reference editing up to 2K.

Microsoft AI's image model for photorealistic and artistic images from a prompt or one reference image, in seven aspect ratios.

Google's fastest, most cost-efficient Nano Banana model — 1K images in ~4 seconds for high-volume workflows.

Recraft's design-focused image model, built around brand-consistent typography, layout, and colour control.

Recraft's premium design tier, for higher-resolution print-ready marketing and packaging visuals.

OpenAI's state-of-the-art image generation model with precise instruction-following for both text-to-image and image editing.

Recraft's vector model, generating clean scalable SVG artwork with editable paths for logos, icons, and illustration.

Recraft's premium vector tier, producing higher-fidelity SVG artwork for branding and production layout work.

Ultra-fast image generation model optimised for speed and creative output.

Lightweight Seedream variant for fast, high-quality image generation.

Black Forest Labs' top-tier Flux 2 model — highest fidelity, strongest prompt adherence, and most stable multi-reference editing.

ByteDance's versatile image model with strong aesthetic quality and prompt fidelity.

Google's Gemini 3 Pro image model, built for accurate text rendering, multi-image blending and studio-grade composition at up to 4K.

Professional-grade Flux model for photorealistic and artistic image synthesis.

Google's original Nano Banana, the Gemini 2.5 image model for fast generation and conversational editing that holds a subject's likeness across successive edits.

Mistral's compact transcription model, a lower-cost sibling of Voxtral Small Transcribe 3 with timestamps and language detection. In public preview.

Mistral's flagship transcription model, which returns a timestamped transcript and detects the language on its own. In public preview.

Mistral's text-to-speech model with preset voices in eight languages; emotion and pace tags written into the text, such as [emotion=happy] or [pace=slow], direct the delivery. In public preview.

A lower-latency version of HeyGen Voice with the same preset voices on OpenRouter and Runware, and voice cloning on Runware.

HeyGen's text-to-speech model with preset voices on OpenRouter and Runware; on Runware it can also clone a voice from a recording.

AssemblyAI's speech-to-text model and the successor to Universal-3.5 Pro, served through its synchronous API for clips of up to two minutes. The app sends a longer recording in two-minute pieces and returns plain text.

Fish Audio's text-to-speech model for directed voice performances: you describe how each line should be delivered in square brackets, add sounds and pauses, and script several speakers in one call.

ElevenLabs' batch transcription model tuned for clinical and medical terminology, which returns word-level timing for subtitles and can label who is speaking.

Microsoft AI's highest-fidelity text-to-speech model, with detailed prosody and a consistent speaker over long passages. It reads in 23 languages through preset voices across 28 locales, and the voice you pick sets the language.

Microsoft AI's low-latency text-to-speech model, built for real-time responsiveness. It reads in 23 languages through the same preset voices as MAI-Voice-2.1, across 28 locales, and the voice you pick sets the language.

Google's speech-to-text model for fast synchronous transcription with automatic language detection; this route returns plain text.

ElevenLabs' most expressive speech model, with closer voice cloning, audio tags for delivery and more than 90 languages.

The low-latency tier of ElevenLabs v4, with the same audio tags and more than 90 languages at half the per-character price.

Fish Audio's speech-to-text model for multi-speaker conversations, which labels who is speaking and keeps emotion and vocal-event cues such as laughter in the transcript.

Google's most expressive Gemini speech model, for long-form narration and two-speaker scenes in 130 languages, with inline vocal cues such as sighs and laughs.

A faster, lower-cost Gemini speech model for high-volume narration and voice agents in 101 languages, with the same voices and vocal cues as 3.8 Flash TTS.

Meta's speech-to-text model for push-to-talk and speaker-aware transcription, which returns plain text.

Microsoft AI's multilingual transcription model, which covers 60 languages with automatic language detection and returns plain text.

A variant of ElevenLabs' v3 voice model tuned for natural dialogue in conversational agents, in more than 70 languages.

OpenAI's high-accuracy transcription model for recorded audio, which follows a hint for names and terms and returns plain text without timestamps.

Fish Audio's speech-to-text model, which detects the spoken language automatically; timestamps come back on the direct route.

Fish Audio's recommended production TTS with natural-language cues in brackets, multi-speaker synthesis in one call, and zero-shot voice cloning across 83 languages.

ByteDance's audio generation model that reads a script word for word or builds a whole scene from one prompt, with a described voice, ambience, music and sound effects, and clones a voice from short reference clips.

Fish Audio's previous-generation open S2 TTS with natural-language cues in brackets, multi-speaker dialogue and zero-shot voice cloning across 80+ languages.

Google's professional-grade music model with enhanced fidelity and musical expression.

ElevenLabs' most expressive voice model with rich emotional and tonal range.

Enhanced Lyria model with improved musical coherence and audio quality.

ElevenLabs' batch transcription model, which returns word-level timing for subtitles and can label who is speaking.

OpenAI's generally available speech model: a chat model that reads text aloud in 13 natural-sounding voices, with consistent voice identity across long passages.

The cost-efficient tier of OpenAI's speech model, reading text aloud in the same 13 voices at a fraction of the price.

A lower-cost version of GPT-4o Transcribe for high-volume work that returns plain text without timestamps.

OpenAI's most accurate hosted transcription model, which follows a hint for names and terms and returns plain text without timestamps.

ElevenLabs' English-only speech model with very low latency, suited to conversational and real-time use.

ElevenLabs' fastest speech model, with about 75 ms latency across 32 languages at half the per-character price of Multilingual v2.

A faster, slimmed-down Whisper Large V3 that keeps most of its accuracy and returns timed segments for subtitles.

OpenAI's open-weight Whisper Large V3, which transcribes speech in close to a hundred languages and returns timed segments for subtitles.

ElevenLabs' previous-generation voice model for lifelike, consistent long-form narration such as voice-overs and audiobooks, in 29 languages.

OpenAI's hosted Whisper model, which transcribes speech in dozens of languages and returns timed segments for subtitles.

xAI's low-cost Grok Imagine video tier: clips of 1 to 15 seconds with synchronized audio from text or a starting image, at 480p to 1080p.

HeyGen's general-purpose video model: clips of 5 to 15 seconds with synthesized dialogue, ambience and sound effects, from text, a starting image or reference media, at 480p or 768p.

MiniMax's throughput-tuned H3, built for faster clips with native synchronized audio and first-to-last frame control. It trades H3's multi-reference conditioning and 2K tier for speed.

Alibaba's lower-latency Wan 3.0 tier, with the same multimodal workflows, 30-second ceiling, and native audio as the base model.

Alibaba's all-in-one Wan 3.0 video model, generating clips of up to 30 seconds with native audio from text, frame images, or a deep bench of reference media.

ByteDance's flagship multimodal video model — native synchronized audio, clips up to 30 seconds, and large reference sets for character and scene consistency.

Black Forest Labs' multimodal video model: text, image, or video in, with synchronized audio, lip-sync, and a cheap draft mode.

MiniMax's Hailuo 3 video model with native synchronized audio, multi-reference consistency, and prompt-driven editing of a finished clip.

Alibaba's video model with three modes — text-to-video, image animation, and reference-to-video from up to 9 images.

ByteDance's lightweight Seedance 2.0 tier — about half the cost per second, 480p/720p with synchronized audio.

xAI's image-to-video model — animates a still image into a short clip with synchronized audio.

Google's cost-efficient Veo tier, generating short clips with native audio at roughly half the price of Veo 3.1 Fast.

ByteDance's low-latency Seedance 2.0 tier — quick 480p/720p generation with synchronized audio.

Kuaishou's cinematic video model generating clips up to 1080p with native audio, lip sync, and sound effects from text or images.

ByteDance's advanced video model with improved motion coherence and scene quality.

Runway's flagship video model with realistic physics and strong visual fidelity for text-to-video and image-to-video.

Google's high-quality video model with native audio, reference-image support, and frame-to-frame control.

Google's fast video model for generating high-quality short clips from text prompts.

Nace.AI's largest decision model, the successor to Drex v2.0, with a 262K-token context. Its host takes 16 questions per request, so the app sends a larger set in batches.

Perplexity's 27B decision model, the successor to Decider V1.1 with the same API: it reads text and images and answers typed questions with calibrated probabilities, choices and scores.

Respan's open-weight decision model, built on Gemma 4 26B-A4B: it reads text and images and answers the questions you define with a choice, a score or a yes/no probability.

Liquid AI's largest decision model, served as a System One endpoint: it reads text and images and answers typed questions with a choice, a score or a yes/no probability.

Upstage's low-latency decision model, a faster variant of Solar Decide built on Solar Mini 4, with a context window of over 500K tokens.

Nace.AI's 35B mixture-of-experts decision model with 3B active parameters, the larger successor to Drex v1.5. Its host takes 16 questions per request, so the app sends a larger set in batches.

Microsoft's small decision model, which reads the content it is given and returns a calibrated probability for each fixed answer option instead of writing text.

The mixture-of-experts member of Cloudflare's open-source Clef decision family, a fine-tune of Qwen3-Omni 30B with 3B active parameters that reads text and images and returns typed decisions.

GPT 6 Luna served through OpenAI's Decisions API: it reads a state and returns typed, probabilistic answers to choice, score and yes/no questions instead of writing text.

TypeSafe AI's System One decision model: it answers choice, yes/no and score questions about a text with probabilities, and never writes free text.

Nace.AI's 9B decision model, which reads a text or JSON state and returns a calibrated probability for each answer to yes/no, multiple-choice and ordinal questions.

Cloudflare's open-source 27B decision model, fine-tuned from Qwen3.8 27B to turn a state and a schema of typed questions into decisions.

The fast 9B member of Cloudflare's open-source Clef decision models, fine-tuned from Qwen3.5 9B for typed classification and routing.

Together AI's experimental decision model, fine-tuned from Qwen3.5 4B for routing, triage and policy checks, with open weights.

Liquid's structured decision model, served as a System One endpoint for choice, score and yes/no answers.

Upstage's decision model with a context window of over 500K tokens, for classifying long documents in one pass.