August 2025 Summaries

OpenRouter has unveiled its first image model, Gemini 2.5 Flash Image Preview, which merges the cognitive capabilities of large language models (LLMs) with the visual finesse of diffusion models. This innovative model facilitates various workflows, such as generating website pages, crafting comic panels with consistent characters, and producing numerous logo variations efficiently. It boasts advanced features like state-of-the-art image generation, exceptional character consistency across scenes, and the ability to generate multiple images simultaneously with a 32k context length, all at a cost of $0.30 per million input tokens. OpenRouter has partnered with Google DeepMind for this launch, and further insights are available in a blog post by Google DeepMind.

OpenRouter has introduced the Gemini 2.5 Flash Image Preview, marking its first-ever image model that integrates the capabilities of large language models (LLMs) with the visual finesse of diffusion models. This innovative model facilitates various workflows, such as generating website pages, crafting comic panels with consistent characters, and creating bulk logo iterations. Key features include state-of-the-art image generation and editing, remarkable character consistency across scenes, and the ability to produce multiple images in a single call, with a 32k context length. The service is available for $0.30 per million input tokens, and OpenRouter is collaborating with Google DeepMind as a launch partner. Users are encouraged to follow OpenRouter on X for updates on model features.

GPT-5, OpenAI's latest model, has been launched with significant improvements in reasoning, reliability, and specialized applications, making it highly effective for complex tasks, including health-related inquiries and code-intensive challenges. The model introduces "dynamic reasoning" to deliver expert-level responses, reducing hallucinations by approximately 80% when using its "thinking mode." It supports an extensive context window of 400,000 tokens and allows up to 128,000 output tokens. Available in multiple variants such as GPT-5, GPT-5-mini, and GPT-5-nano, it caters to different needs from complex reasoning to high-throughput tasks, each with distinct pricing options. The model's performance is substantiated by SWE-Bench verification, where GPT-5 achieved an accuracy of 74.9%, the highest among its peers. OpenAI has retired the Horizon Alpha and Beta endpoints, making GPT-5 accessible via API with a requirement for users to bring their own API key.

OpenAI's GPT-5, the latest and most advanced model from the company, introduces significant improvements in reasoning capabilities, reliability, and domain-specific applications, notably excelling in complex reasoning and health-related inquiries. The model features a dynamic reasoning ability, which greatly reduces the incidence of hallucinations and supports a vast 400,000 token context window, accommodating up to 128,000 output tokens. It's available in different variants, including GPT-5-mini and GPT-5-nano, each tailored for specific use cases ranging from balanced performance and cost-efficiency to high-throughput tasks. Pricing varies according to the variant, with the standard GPT-5 model commanding higher costs per million tokens compared to its mini and nano counterparts. Earlier versions, known as Horizon Alpha and Beta, have been phased out, and users are required to use their own API key to access GPT-5 through OpenAI's API.

OpenRouter has introduced new features that enhance app interaction by allowing users to input audio files in formats like wav or mp3 and send PDF links directly to OpenRouter without the need for base64 encoding. These updates aim to streamline the integration of audio and document workflows into applications, making it more convenient for developers to implement these functionalities with any language model on OpenRouter. The provision of guides and documentation further supports users in leveraging these features effectively, thereby simplifying the process of adding sophisticated input methods to their apps.

OpenRouter has introduced new features that enhance app functionality by allowing audio and PDF URL inputs. Users can now interact with apps using wav or mp3 audio files, with models available to support these audio inputs, as detailed in the Audio Input Guide. Additionally, PDF URLs can be directly sent to OpenRouter without the need for base64 encoding, simplifying the integration of document workflows with any LLM on OpenRouter. The PDF URL Docs provide guidance on leveraging this feature, making it easier to incorporate document-related processes into applications.