Agentic AI Platform Change Log

Observability / auditability

OpenRouter introduced a new analytics dashboard and Analytics API for tracking AI usage across all agents, models, and requests. The update allows users to monitor spend per model, save custom charts, click through to underlying logs, and query usage data directly from the terminal.

Engineering and finance teams can now closely monitor and attribute AI costs across complex multi-model deployments. The programmatic API access ensures usage data can be ingested into internal enterprise observability tools.

Funding / partnership

OpenRouter announced it is joining forces with Stripe in an acquisition reportedly valued at over $7 billion. The deal integrates OpenRouter's unified AI model routing platform with Stripe's broader infrastructure ecosystem.

Enterprise buyers gain assurance of long-term platform stability and backing from a major infrastructure provider. This mitigates the typical vendor risk associated with startups and paves the way for deeper billing and payment integrations.

MCP / tool calling / API

OpenRouter released Ori Harness, a new CLI tool that allows developers to run coding agents like Claude Code, Codex, OpenCode, and Hermes. The tool provides an optimized configuration for integrating these agents directly with OpenRouter's model routing infrastructure out of the box.

Engineering teams can significantly reduce the setup time required to connect popular autonomous coding agents to OpenRouter. This streamlines the deployment of agentic workflows by eliminating manual configuration and standardizing the connection layer.

Observability / auditability

OpenRouter launched Classifiers in beta, enabling developers to automatically tag generation requests with structured metadata. Users can define custom taxonomies, such as task type or compliance category, and use a small model like Gemini 3.5 Flash Lite to tag the traffic. The resulting tags are aggregated and integrated directly into the Activity Explorer for usage reporting.

Teams gain powerful, built-in observability to track exactly what their AI agents are doing, which models they use for specific tasks, and where spend is accumulating. This eliminates the need to build and maintain custom logging, classification, and attribution pipelines.

MCP / tool calling / API

OpenRouter introduced a dedicated audio transcription endpoint (/api/v1/audio/transcriptions) that accepts base64-encoded audio. The service routes requests to Whisper-class and newer speech-to-text models across multiple providers, offering automatic load balancing and delivering JSON transcripts alongside usage statistics.

Developers can natively convert speech to text using their existing OpenRouter setup and API keys, completely avoiding the operational overhead of running standalone Whisper servers or integrating secondary SDKs. This centralizes billing and authentication for text and multimodal agent workflows.

Agent capability

OpenRouter detailed a new architecture pairing prompt caching with sticky routing. By passing a stable session_id in API requests, multi-turn agent conversations are automatically routed back to the specific provider holding a warm cache of the prompt.

Agents that rely on large, repetitive system prompts or intricate tool schemas can drastically reduce inference costs. Developers simply include a session_id in their payloads to ensure follow-up requests hit cached tokens, which are priced at roughly 10% to 50% of the cost of fresh inputs.

Agent capability

OpenRouter released a unified API endpoint that supports multiple AI modalities, including text, image generation, video, text-to-speech, speech-to-text, and embeddings. Developers can switch between models and modalities by changing the model string and content type in their request to a single OpenAI-compatible base URL.

This update significantly reduces the integration overhead for engineering teams building multi-modal applications. Instead of managing separate SDKs, authentication schemes, and billing for different providers, organizations can centralize their AI access. It also enables consistent routing controls and fallback logic across all modalities.