API Market

OpenAI flagship chat model tuned for complex multi-step reasoning and reliable instruction following.

Anthropic's most capable Claude, built for demanding reasoning and agentic workflows.

OpenAI GPT-6 class chat model balancing reasoning depth and response speed.

xAI's Grok chat model with broad world knowledge and direct, concise answers.

The latest M-series language model from MiniMax, suited for Agent reasoning, tool calling, coding and long-context tasks.

DeepSeek high-throughput chat model with solid reasoning at flash-tier latency.

A frontier general-purpose model built for long-horizon coding, complex knowledge work, visual reasoning, and multi-step planning with verification.

Anthropic Claude Fable 5.1 flagship iteration.

A model built for reliable multi-step agentic coding, visual understanding, precise instruction following, and sustained reasoning.

An Opus model for complex reasoning, agentic coding, and specialized knowledge work; the page notes platform support for up to 1M context.

An agentic model built for end-to-end coding, complex reasoning, visual analysis, and long-horizon enterprise work.

A chat model with strong reasoning, coding, native computer use, and adaptive thinking; the page notes platform support for up to 1M context.

A chat model with agentic ability, multimodal input, native computer use, and self-verification.

Anthropic Claude Sonnet 5.5 balanced chat model.

The Flash edition of the DeepSeek-V4 series with substantially improved agent capability, benchmarking well above V4-Pro-Preview; a cost-efficient variant that still supports a 1M-token context window with strong reasoning at lower cost and faster response, close to Pro-level performance on simple-to-medium agent tasks. Suited to conversation, content creation, and lightweight agent calls, with switchable thinking and standard modes.

DeepSeek's flagship model, built around a million-token context window and long-form output, with built-in thinking mode suited to complex agent tasks, long documents, and multi-turn reasoning.

The lite tier of ByteDance's Doubao-Seed-2.0 family, tuned for low-latency, high-concurrency, cost-sensitive workloads while keeping the family's multimodal understanding, long-context reasoning, and tool-augmented execution; supports four levels of thinking depth.

The mini tier of ByteDance's Doubao-Seed-2.0 family, built for high-concurrency, low-latency, cost-sensitive workloads while keeping the family's multimodal understanding, long-context reasoning, and tool-augmented execution; a lightweight option for general production use.

A preview release aimed at improving on Gemini 3 Pro with stronger reasoning, better token efficiency, and more grounded, factually consistent responses; optimized for software engineering behavior and usability, and for agentic workflows that need precise real-world tool use and reliable multi-step execution.

Google's fastest and most cost-efficient model in the 3.5 series, built for low-latency, high-throughput agent tasks with adjustable thinking levels, up to 350 tokens/sec output and a 64,000-token max output.

An OpenAI Chat Completions-compatible calling variant of Gemini 3.7 Flash.

Gemini 3.8 Flash cost-efficient chat model.

Google Gemini 4 Argon flagship chat model.

A low-cost, low-latency model suited to chaining or parallelizing multiple model calls (such as calling several APIs), passing large context to the model (such as full codebases or conversation history), and fast, real-time text interactions like customer support chatbots.

A reasoning model built for agentic coding, knowledge work, research, and complex multi-step tasks.

The Luna tier of the GPT-5.6 series; the detail page also lists Luna, Terra, and Sol together. This entry covers the Luna calling model, supporting mixed text, image, and file input, adjustable reasoning effort, and tool calling.

The flagship tier of the GPT-5.6 series, suited to complex professional work; roughly corresponds to the unsuffixed tier in the earlier GPT-5 series.

A GPT-5.6 tier built to balance intelligence and cost; roughly corresponds to the mini tier in the earlier GPT-5 series.

OpenAI GPT-6 Luna low-cost workhorse tier.

A frontier reasoning model from xAI built for long-horizon agents, complex reasoning, code, and knowledge work; supports structured multimodal input, adjustable reasoning effort, web search, and function calling.

Kimi's flagship model, built for long-horizon programming and end-to-end knowledge work, with a 1M-token context window, native multimodal support, and leading overall intelligence.

A compact model in OpenAI's o-series of reasoning models, focused on cost efficiency and speed, with strong performance on math, coding, and vision tasks, suited to everyday generation and analysis.

Alibaba Qwen's flagship model built for the agent era, ranking first among domestic models on global blind-test leaderboards; leads or ranks in the first tier globally on coding agents, general agents, and reasoning, with dual thinking/non-thinking modes, a 1M-token context window, and the ability to execute autonomously for up to 35 hours across frameworks.

A 2.4-trillion-parameter MoE flagship model with major gains in coding and office-work capability, able to autonomously develop and deliver complete projects over many days; handles hundreds of specialized tasks across law, finance, and design in a single end-to-end conversation. Native visual understanding runs through planning, execution, and verification, with deep semantic parsing of long documents and long video, plus autonomous planning and iterative refinement on long-running tasks.

GPT-Image 2.5 variant tuned for vivid, high-contrast renders with reliable prompt adherence.

GPT-Image 2.5 variant tuned for warm, luminous tones and clean composition.

Cost-optimized GPT-Image 2.5 variant, same family quality at lower credits.

Google's Nano Banana 2.1 image model, strong at prompt adherence, text rendering and edits.

Cost-optimized sunburst variant for warm-toned images at lower credits.

ByteDance Seedream 5.0 text-to-image model with excellent typography and photorealistic detail.

A Pro text-to-image variant in the FLUX.2 image model suite, focused on realistic detail, multi-reference consistency, and text rendering; the page also lists Pro image-to-image, Flex text-to-image, and Flex image-to-image.

A contextual image generation and editing model that combines text and image input for precise, consistent results.

A proven text-to-image baseline model for high-quality visuals, pairing fast inference with strong prompt adherence for scalable text-based image generation.

An image generation model from OpenAI, with entry points for text-to-image and image-to-image on its detail page.

OpenAI GPT Image 2.5 (sunburst, high fidelity). 1K/2K/4K, 13 aspect ratios.

The third generation of Ideogram image models, offering text-to-image, editing, and Remix, with Reframe also noted on the page; focused on text layout, image consistency, and creative control.

A next-generation Midjourney V7 image model with stronger prompt understanding and consistency, improved realism, detail, and style control, plus faster and more stable generation; this variant covers fast-mode text-to-image.

Google Gemini 3.1 Flash Image model supporting fast 1K, 2K, and 4K image generation along with image editing.

Google Gemini 3.1 Flash-Lite Image model built for fast 1K image generation and prompt-based image editing.

A high-fidelity image model from Google DeepMind built on Gemini 3.0 Pro Image, supporting 1K, 2K, and 4K generation, text rendering, and character consistency.

A multimodal image model from ByteDance combining text-to-image, image-to-image editing, and multi-image output, with improved editing consistency, multi-image blending, and finer control over image detail, including more natural rendering of small text and faces.

The Lite tier of the Seedream 5.0 family, a unified multimodal image model built for faster generation and lower cost in text-to-image workflows; this entry covers the default text-to-image variant, with an image-to-image editing variant also available on the page.

A multimodal image model whose default page variant is image-to-image, built for controllable generation, precise editing, complex information visualization, realistic portraits, and multilingual visual creation.

MiniMax H3 is a versatile omni-modal generation system. It jointly understands multimodal context of text, images, video and audio, and generates 768P/2K videos up to 15 seconds with native stereo audio (dialogue, music and sound effects can be described in the prompt).

MiniMax H3 is a versatile omni-modal generation system. It jointly understands multimodal context of text, images, video and audio, and generates 768P/2K videos up to 15 seconds with native stereo audio (dialogue, music and sound effects can be described in the prompt).

MiniMax H3 is a versatile omni-modal generation system. It jointly understands multimodal context of text, images, video and audio, and generates 768P/2K videos up to 15 seconds with native stereo audio (dialogue, music and sound effects can be described in the prompt).

Kling 3.0 Turbo text-to-video: fast, cost-efficient generation at 720p/1080p, 3–15 seconds.

Kling O3 image-to-video: animate first/last frame images with native audio and multi-shot narrative, up to 4K.

Kling O3 reference-to-video: keep characters and scenes consistent with up to 4 reference images, native audio, up to 4K. Reference video input is not enabled yet.

Kling Video O1 (standard tier): text-to-video and image-to-video, 3–10 seconds.

ByteDance Seedance 2.0 fast-tier video model, trading a little fidelity for quicker renders.

A video model supporting multimodal references, native audio, and multi-shot narrative, with clips up to 15 seconds.

Built on a unified multimodal training framework, this Kling 3.0 series model delivers more native multimodal input and output, combining synchronized audio-visual generation with subject consistency while removing prior duration limits. It supports longer video generation (up to 15 seconds) with native audio-visual output, flexible shot control, and precise semantic response, upgrading from the earlier Kling-Video-O1 and Kling-V2-6 with markedly improved realism and more expressive character performance.

A cost-efficient model in the Kling 3.0 video series, with image-to-video as the default variant.

A video generation model built as a full creative workflow rather than standalone clip generation, supporting single-generation clips up to 15 seconds at 1080p, multi-shot storytelling with native synchronized audio-video, precise camera control, cross-frame character consistency, and multilingual on-frame text rendering; can be integrated into agent workflows via a CLI interface for creative and commercial-scale production.

A multimodal video model from ByteDance that combines text, image, video, and audio references to generate coherent video with multi-shot consistency and native audio.

A multimodal video model built for faster generation and lower inference cost, supporting text-to-video, image-to-video, consistent characters, and cinematic camera movement.

A generation model supporting videos up to 30 seconds, multimodal references, precise local editing, and native 4K output.

Tongyi Wanxiang 3.0 video generation supporting text-to-video and image-to-video, up to 30s at 1080P with optional audio.

Higher-fidelity Prime tier of Wan 3.0 video generation; same parameters as the standard tier.

An AI music generation model that composes full songs from text prompts and style tags; the V5.5 release emphasizes expressive vocals and personalized songwriting based on a user's own voice, original tracks, and musical preferences.

Uses the official Suno stem-separation API to split a generated track into clean vocal, backing, or individual instrument stems using advanced source separation.