AI Open Models

OpenAI-compatible LLM inference and chat completions across seven open-weight reasoning, coding, and vision models. Choose Kimi K3, Qwen3.6, Qwen3.8, GLM-5.2, GLM-5.3, GLM-5.3-Flash, or DeepSeek V4.1 Flash. Use them to generate text or JSON, analyze images, and call client-provided tools. Supports up to 4 images and selectable reasoning controls. Pricing is quoted dynamically from the prompt and max_tokens.

Routes requests across open-weight models using task fit, tool calling, vision, JSON format, context, output capacity, provider availability, and price. Send POST application/json with model set to ai-open-models/auto and a non-empty messages array. task_type is optional; when omitted, Auto classifies the request. Supports function tools, JSON responses, and mixed text/image messages using public HTTPS image URLs or base64 data URIs. Multipart uploads and streaming are not supported.

Fusion Model. Internal reviewers examine the conversation and available tool results, then a final model returns the next OpenAI-compatible response, including tool calls when tools are provided. Supports repeated multi-turn calls, client-executed tools, and JSON responses. Internal review notes are not returned.

Run the same task against several explicitly selected open-weight models in parallel and return every answer separately. No ranking, verification, or synthesis.

Retrieve clean content from one webpage using Exa. Choose full text, query-focused highlights, or an AI-generated summary, with page metadata and retrieval status.

Converts one natural-language task into validated OpenAI-compatible JSON request bodies for multiple open-weight models. Uses the optional models list when supplied; otherwise uses all catalogue models compatible with the input. Supports text, visual input, and model-supported reasoning levels. Returns JSON request bodies only and does not execute inference.

POST application/json with required non-empty query. Exa-powered search returns ranked URLs and query-relevant highlights. Optional fields are objective, num_results (default 5), category, include_domains, exclude_domains, start_published_date, and end_published_date.

AI-generated answers powered by Exa web search. Ask a factual or open-ended question and receive a synthesized answer with source citations.

Conduct one live-web research task with 2-7 buyer-selected open-weight analyst models, then use a buyer-selected synthesis model to produce one structured, cited research report. Reconciles agreements, disagreements, contradictions, and uncertainties while preserving sources, analyst outputs, latency, usage, and validation results.

An LLM router for an open-weight inference service. It analyzes your request, selects a suitable compatible model, runs the selected model, and returns its answer in OpenAI-compatible format.

Run one live-web research task through 2–7 buyer-selected open-weight models. Every model receives the same retrieved evidence and returns a separate structured analysis with citations, latency, usage, and validation results. No ranking or synthesis.