WaveSpeed launches a unified LLM API granting access to more than 260 language models, including major players like GPT, Claude, Gemini, Grok, DeepSeek, Llama, Qwen and Mistral, plus over 1,000 AI models for multimodal generation across image, video, audio, 3D, and avatars.

The WaveSpeed LLM API is live at wavespeed.ai/llm, offering a free playground and clear pricing and documentation at wavespeed.ai/models, wavespeed.ai/docs, and wavespeed.ai/pricing.

WaveSpeed positions itself as a platform to access 1,000+ AI models via a single API key with transparent pricing, covering language, image, video, audio, 3D, and avatars.

The platform supports multimodal workflows, enabling a single workflow to combine reasoning, image generation, video creation, and speech output to reduce infrastructure complexity.

Pricing is per-token, transparent, designed to minimize cold starts and latency, with no long-term commitments.

Emphasis on low cold-start latency, separate input/output per-token pricing, and no subscriptions or minimum commitments.

The API allows developers to evaluate, route, and deploy multiple LLMs without juggling separate SDKs, API keys, billing systems, or model-access workflows, all in one integration and billing stream.

Developers can compare, switch, and deploy multiple models via a single API key and billing account, reducing the need for multiple SDKs and keys.

The API uses a standard chat-completions interface with streaming, tool usage, vision input, and JSON mode, accessible via common HTTP clients and SDKs, and supports per-model pricing with no subscriptions.

One endpoint handles chat-completions with streaming, tools, vision, and JSON modes; pricing is per-model with no subscriptions or minimum commitments.

Model catalog features Claude Opus 4.7, Claude Sonnet 4.6, GPT-5.5, Gemini 3, Grok 4, DeepSeek V4, Llama 4, and Qwen 3, with side-by-side comparisons by price, context window, and capability tags.

Model discovery supports side-by-side comparisons by price, context window, and capability tags (vision, audio, tool use), and lets users switch models by changing a single parameter.