July 2025 Summaries

Presets introduced by OpenRouter allow users to effortlessly transfer model configurations across different applications, optimizing efficiency by setting up preferred settings once and reusing them as needed. Users can create a preset by visiting the OpenRouter website, where they can customize various settings including default model prompts and sampling parameters. After configuring, users can test the preset to ensure it functions as desired, and then apply it within any app that supports custom models. Additionally, the ability to revert to previous versions using version history provides flexibility and ease in managing configurations.

Presets offer a streamlined solution for transferring model configurations across various apps, allowing users to set up their preferred settings once and reuse them to save time and effort. The process involves creating a preset through a designated website, customizing settings such as the default model, system prompts, and sampling parameters, and testing to ensure it functions as desired. Once configured, these presets can be integrated into any application that supports custom models, with the flexibility to select the underlying model using the preset field. Additionally, users can revert to previous configurations using version history, providing a safeguard against unwanted changes.

Venice is a new privacy-focused provider offering open-source AI models designed to uphold user autonomy, privacy, and creative freedom, with their featured model being the Dolphin Mistral 24B Venice Edition. This model, based on Mistral 24B and fine-tuned by Venice, provides a notably uncensored AI experience with a low censorship refusal rate of 2.2%, significantly lower than that of Claude and GPT-4o-mini. It ensures complete privacy by not storing or filtering data server-side and is freely accessible through OpenRouter's API and Chatroom.

Venice has launched a new privacy-focused open-source AI model, the Dolphin Mistral 24B Venice Edition, which emphasizes user autonomy, privacy, and creative freedom. This model is distinguished by its notably low censorship refusal rate of 2.2%, significantly lower than other models like Claude and GPT-4o-mini, with rates of 71% and 64%, respectively. The Dolphin Mistral 24B is fully private, ensuring no server-side storage or filtering, and is accessible for free through OpenRouter’s API and chatroom. Venice's initiative underscores its commitment to providing one of the most uncensored AI experiences currently available, reflecting a partnership aimed at enhancing privacy and user control.

Users of the AI coding assistant Cursor can now enhance their experience by integrating it with OpenRouter to access Moonshot AI's Kimi K2 model, which boasts 1 trillion parameters and a leading score in code generation capabilities. This integration is straightforward, requiring configuration of API keys and model selection within Cursor's settings. Despite a potential warning about limited agentic support, Kimi K2 is fully equipped for assistant-style workflows. Users are encouraged to obtain their API key from OpenRouter and can explore available models and join the developer community through OpenRouter's platform.

Moonshot AI's Kimi K2 model, now available on the OpenRouter platform, can be integrated into Cursor, an AI coding assistant, to enhance code generation and tool usage. With 1 trillion parameters and a 65.8% SWE-Bench Verified score, Kimi K2 is leading in open-source code generation models. To use it, users need to configure Cursor settings by activating the OpenAI API key, pasting the OpenRouter API key, and altering the OpenAI Base URL to OpenRouter's URL. After adding and enabling the model in Cursor, users can select Kimi K2 from the model dropdown in the chat pane, ignoring any warnings about limited agentic support, as it remains fully functional in assistant-style workflows. Users can obtain an API key by signing up on OpenRouter's website, which also provides access to other available models and a developer community on Discord.

OpenRouter has announced updates to its free tier models in response to two major providers transitioning to paid-only models. Committed to keeping AI accessible, OpenRouter is expanding its free model capacity by onboarding new providers and covering some costs to maintain popular models' free accessibility. A new model, Venice Uncensored, is now available for free, and community favorites like DeepSeek V3 and DeepSeek R1 will remain free due to existing partnerships. However, some less popular models will no longer be available for free. OpenRouter emphasizes its dedication to sustaining free-tier access and expanding innovative AI tools for its users.

OpenRouter is committed to maintaining free AI inference options despite the transition of major providers to paid-only models, ensuring continued access to AI tools for users worldwide. They are expanding free capacity by onboarding new providers and covering some costs to keep popular models like DeepSeek V3 and DeepSeek R1 freely accessible. OpenRouter has also introduced a new free model, Venice Uncensored, to enhance their offerings. While some less popular models will move away from the free tier, the company emphasizes the importance of sustaining free access to support experimentation, learning, and innovation, and appreciates the community's ongoing support.

Cypher Alpha is a new, versatile model released by Cypher Labs, a fictional provider, and is featured on OpenRouter as of July 1, 2025. The model is equipped with native tool-calling capabilities and is available for free experimentation, though it is subject to OpenRouter's free usage limits. Users interested in trying out Cypher Alpha can access it via an API or chatroom, with the option to engage in discussions and provide feedback through a dedicated model thread on Discord. It is important to note that prompts and completions are logged by the model creator and may be used for training purposes.

Cypher Alpha, a new versatile model released by the fictional Cypher Labs, offers native tool-calling capabilities and is available for free experimentation within OpenRouter's usage limits. Users can interact with the model via API or Chatroom, and the prompts and completions are logged for potential training purposes. Feedback and discussions about Cypher Alpha can be shared on the specified Discord channel, encouraging community engagement and collaborative development.