Open-weight models 4

Low-cost, scalable inference for high-volume AI workloads Doubleword is an AI inference platform for high-volume workloads, offering OpenAI-compatible access to open-weight language, vision, OCR, and embedding models. It supports real-time, async, and approximately 24-hour batch inference, allowing users to trade latency for substantially lower token costs. Doubleword is designed for agentic reasoning, evaluations, synthetic data generation, annotation, extraction, coding workflows, and large-scale data processing. Its inference stack includes optimization technologies such as FlashOffload, Speculative KV Coding, and Cloudburst to improve throughput, cache efficiency, and startup times.

Managed AI infrastructure for open models, agents, and scalable private deployments FlexAI is an agent-native AI infrastructure platform that provides managed inference for more than 20 open-weight models through one OpenAI-compatible API key. It supports text, vision, code, reasoning, embeddings, speech, audio, and image-generation workloads. Users can start with serverless model access, then scale to dedicated GPU endpoints, fine-tuning, and private AI cloud deployments across VPC, on-premises, or air-gapped environments. FlexAI also offers agent tools such as tool calling, streaming, structured outputs, approvals, governance, and audit trails.

Demo platform for OpenAI's open-weight models for developers. The gpt-oss playground is a demo platform designed for developers to explore OpenAI’s open-weight models, specifically gpt-oss-120b and gpt-oss-20b. It allows users to interact with these models and observe their reasoning capabilities.

Open-source database for comparing AI models, providers, specifications, and pricing Models.dev is a comprehensive open-source database of AI model specifications, pricing, providers, capabilities, context limits, output limits, release dates, and update dates. It provides provider-agnostic model metadata, lists the providers serving each model, and organizes canonical models by laboratory or author. Users can compare models such as GPT-6 Astra across multiple providers, review features including reasoning, tool calling, structured output, temperature support, and weights availability, and access the data through JSON APIs or a type-safe SDK.