The latest AI news and updates of OpenRouter — Unified API for accessing hundreds of LLMs across providers with intelligent routing and optimization. Covering OpenRouter's latest product updates, launches, and analysis from the past 90 days.
OpenRouterJul 27OpenRouter added Moonshot AI’s Kimi K3 multimodal reasoning model to its unified API, offering access through multiple third-party providers. The platform supports reasoning tokens for step-by-step thinking and provides performance metrics for provider selection. Kimi K3 Fast variants from providers including wafer_ai and Fireworks are scheduled for release soon.
OpenRouterJul 27OpenRouter, in partnership with OpenAI, is offering a 50% discount on GPT-5.6 Terra and Luna models. The price reduction applies to input, output, and cache read costs for the first-party OpenAI provider. Pro versions of both models are included in the limited-time offer, which is available exclusively through the OpenRouter API.
OpenRouterJul 25OpenRouter launched a Discover page to explore AI models, featuring live rankings and performance metrics. The page highlights Anthropic’s Opus 5 and OpenAI’s GPT 5.6 Sol as top performers, while providing access to routing solutions and always-latest model aliases. Filters surface models by modality, speed, and cost-efficiency to support model selection.
OpenRouterJul 24OpenRouter added xAI’s Grok STT 1.0 to its unified API. The model supports transcription in 25 languages, word-level timestamps, and speaker diarization. It is available for $0.10 per audio hour.
OpenRouterJul 24OpenRouter added Ant Group’s Ling-3.0-flash, a 124B-parameter Mixture-of-Experts model with 5.1B active parameters per token. Designed for token-efficient agentic inference, the model features a 262,144-token context window and 32,768-token output limit. It is available for free through August 3, 2026.
OpenRouterJul 23OpenRouter now provides access to Alibaba’s Qwen-Audio-3.0-TTS Flash and Plus models through its unified API. These text-to-speech models support 16 languages and accept natural-language direction, including expressive inline cues like [gasp] or [angry]. Flash is optimized for low-latency real-time interaction, while Plus prioritizes voice fidelity and naturalness for polished speech generation.
OpenRouterJul 23OpenRouter added Microsoft’s MAI-Image-2.5-Pro and MAI-Voice-2-Flash to its unified API. MAI-Image-2.5-Pro delivers high-fidelity image generation with precise editing and accurate in-image text rendering. MAI-Voice-2-Flash provides a 2x faster, 15-language voice model optimized for real-time, high-volume experiences.
OpenRouterJul 23OpenRouter has added Deepgram’s Nova-3 speech-to-text and Aura-2 text-to-speech models to its unified API. Nova-3 supports monolingual and multilingual transcription, while Aura-2 provides multilingual text-to-speech across seven languages, including English, Spanish, and Japanese. These models are now available for programmatic access through OpenRouter’s OpenAI-compatible API endpoints.
OpenRouterJul 21OpenRouter saved 22,000 users over $100,000 on GLM 5.2 inference this past weekend. The platform’s automated routing system identified low-demand periods across providers, resulting in price drops of over 50% compared to weekday rates. Users received these savings automatically without needing to adjust their configurations or monitor provider traffic.
OpenRouterJul 20OpenRouter added Krea’s Krea 2 image generation models to its unified API. The family includes Krea 2 Large for textured photorealism, Krea 2 Medium for balanced, cost-efficient illustration, and Krea 2 Medium Turbo for rapid iteration. These models support resolution, aspect ratio, and seed parameters, with Krea 2 Large priced at $0.06 per image.
OpenRouterJul 20OpenRouter added xAI’s Grok Imagine Video 1.5 to its unified API. The model animates still images using text prompts to control motion, camera work, and atmosphere, while generating synchronized sound effects and dialogue. It is available via the platform’s video generation endpoint, priced at $0.08 per second of output and $0.01 per input image.
OpenRouterJul 14OpenRouter updated its MCP server, allowing coding agents to discover, test, and report on models directly within the editor. New features include real-traffic task insights, granular model search filters, and provider pinning for reproducible evals. The update also adds image generation and feedback tools, while renaming tools to a standard verb-noun format for improved usability.
OpenRouterJul 13OpenRouter now treats service tiers as first-class objects, surfacing tier-specific latency and throughput data in the models endpoint API. Requests route using tier-suffixed slugs like openai/flex and xai/priority, or the service_tier parameter. This update provides granular control over cost and latency tradeoffs across supported providers including OpenAI, Google Vertex, and xAI.
OpenRouterJul 8OpenRouter introduced a static model slug, ~x-ai/grok-latest, that automatically routes to the newest Grok model from xAI. The alias currently points to Grok 4.3 and updates to future versions, such as Grok 4.5, the moment they go live. This implementation eliminates the need for manual code changes when new model versions become available.
OpenRouterJul 8OpenRouter introduced a one-click Zero Data Retention toggle within its Chatroom interface. This setting enforces privacy by routing requests only to endpoints that do not retain prompt or completion data. The toggle applies to all models in a comparison session, ensuring consistent data handling across side-by-side testing.
OpenRouterJul 8OpenRouter benchmarked 1,730 visual reasoning questions and found that setting image detail to low often backfires on reasoning models. On gpt-5.5, low detail reduced accuracy by 13.8 points and increased costs by 0.6¢ per question as the model burned extra reasoning tokens. For reasoning models, keeping detail on auto and capping reasoning effort is more cost-effective.
OpenRouterJul 6OpenRouter reports that DeepSeek’s token share on its platform doubled from 9% in January to 18% by June. This growth is driven by agentic workloads, which surpassed human token usage in February. DeepSeek V4 Flash, released in April, now accounts for 70% of the company’s agentic token flow, pushing DeepSeek’s weekly volume to 6.6 trillion tokens.
OpenRouterJun 28OpenRouter launched AutoExacto, a routing system that automatically selects the best-performing inference providers for tool-calling requests. The system re-evaluates providers every five minutes using throughput, tool-call telemetry, and GPQA and TAU-Bench benchmark scores. This update, which is on by default, reduced tool-call error rates by up to 88 percent for models like GLM-5.
OpenRouterJun 28OpenRouter now offers Zero Data Retention for xAI’s Grok 4.3, 4.20, and Build 0.1 models. When ZDR is enabled, neither OpenRouter nor xAI retains prompt data for requests routed to these endpoints. This privacy option is available through account settings, guardrails, or per-request API parameters.
OpenRouterJun 26OpenRouter added fast inference endpoints for Z.ai's GLM 5.2 reasoning model from Wafer AI and Fireworks AI. Setting the model slug to z-ai/glm-5.2:nitro automatically routes requests to the fastest available provider based on live traffic data, eliminating the need for manual provider monitoring.
OpenRouterJun 25OpenRouter now hosts OpenWebUI, providing organizations with a single chat interface and unified billing for over 400 frontier and open-weight models. This integration combines OpenWebUI’s administrative, audit, and document grounding features with OpenRouter’s optimized inference routing and usage tracking. The deployment offers a single procurement path for both the chat experience and model access.
OpenRouterJun 25OpenRouter launched an MCP server that provides AI coding agents with live model intelligence, including benchmarks, pricing, and latency data. The server replaces stale training knowledge with real-time catalog information, allowing agents to test prompts and compare models directly within the CLI. Authentication uses a dedicated key with a seven-day expiry and a ten-dollar default spend cap.
OpenRouterJun 24OpenRouter launched a dedicated Image API providing unified access to over 30 image generation models from 8 providers. The API features typed capability descriptors for model discovery, a normalized request schema, and transparent per-request pricing in USD. It also supports provider-specific parameter passthrough and native streaming previews for OpenAI GPT Image models.
OpenRouterJun 24OpenRouter added Z.ai’s GLM 5.2 reasoning model to its unified API, consolidating access across 22 independent providers. The model features a 1-million-token context window and supports high-effort reasoning for complex agentic tasks. One provider currently serves the model at over 125 tokens per second, allowing requests to be routed based on specific speed or pricing requirements.
OpenRouterJun 19OpenRouter introduced workspace budgets that cap inference spending with daily, weekly, monthly, or lifetime limits. The system automatically blocks requests with a 403 error once a limit is reached. These budgets are available on the Enterprise plan and require strictly decreasing limits as the interval narrows.
OpenRouterJun 17OpenRouter launched a Cost Simulator tool that estimates potential savings by comparing model pricing against a user's actual generation history. The tool uses median endpoint pricing to project costs, showing users the impact of switching models. A simulation of 100 generations on Claude Fable 5, for instance, projected a 73.2% cost reduction when switching to GLM 5.2.
OpenRouterJun 15OpenRouter added free API capacity for OpenAI's gpt-oss-20b and Google DeepMind's Gemma 4 26B models, served by provider Darkbloom. Both models are now available at zero cost for input and output tokens through OpenRouter's free-variant endpoints. This update provides cost-free access to these open-weight models for testing and experimentation.
OpenRouterJun 15OpenRouter launched the Fusion API, a compound model that runs multiple LLMs in parallel to synthesize a final answer. On the DRACO deep research benchmark, a panel of budget models achieved performance within 1% of Claude Fable 5 while costing roughly half the price. Fusion is available via the openrouter/fusion slug or as a tool.
OpenRouterJun 13OpenRouter launched the Subagent server tool, allowing models to delegate focused tasks to smaller, cheaper, faster worker models mid-generation. The orchestrator model invokes the tool with a task description, and the worker executes the task independently. Workers can optionally run as sub-agents with their own tools, such as web search, to ground results in fresh sources.
OpenRouterJun 13OpenRouter added a Benchmarks explorer to its rankings page that plots Pareto curves across 10 different benchmarks. It visualizes price-versus-performance trade-offs using Artificial Analysis intelligence indices and Designarena ELO scores for coding, UI, game development, and 3D tasks — surfacing the most cost-effective models for a given use case.
OpenRouterJun 13OpenRouter launched the Activity explorer, a real-time dashboard for tracking AI spending, token usage, and cache performance across models, users, and agents. The tool includes a Trends tab for identifying usage spikes, an Explore tab for granular analytics across any dimension, and a Guardrails tab to monitor security enforcement. It is available now to everyone.
OpenRouterJun 10OpenRouter published guidance showing how to use its Agent SDK's human-in-the-loop primitives to meet incoming AI regulations including the EU AI Act, Colorado's ADMT law, and NIST AI RMF. The guide documents existing SDK hooks that pause execution, persist state, and log human approvals so deployers can gate sensitive agent actions and produce audit trails required by these frameworks.
OpenRouterJun 10OpenRouter has made Anthropic's Claude Fable 5 model available on its platform. This model is designed for complex, long-running coding and autonomous knowledge work, achieving state-of-the-art performance on various benchmarks. Its availability expands access to a frontier AI model for developers building agentic applications.
OpenRouterJun 9OpenRouter introduced its new Advisor server tool, enabling AI models to consult higher-intelligence models during complex tasks. This capability helps prevent models from getting stuck in "doom loops" and allows developers to optimize costs by using expensive reasoning only when necessary.
OpenRouterJun 7OpenRouter now displays real-time cache hit rates and historical traffic data on its Pricing tab. This update provides transparency into how different model providers compare on effective pricing for LLMs like Anthropic's Claude Opus 4.8, enabling users to optimize costs.
OpenRouterJun 5OpenRouter has integrated Sourceful's Riverflow 2.5 image model, which allows users to define a scoring rubric for guiding image generation and adjust reasoning effort. This provides granular control over output quality and composition, enabling more precise alignment with specific creative or brand requirements.
OpenRouterJun 4OpenRouter's automated routing engines now process 13 billion tokens daily, with the coding-specific Pareto Router hitting 1 billion. The milestone coincides with new granular controls that let users manually balance model performance against token costs. This shift highlights how developers are moving from static model selection to dynamic, algorithmic orchestration to manage AI expenses.
OpenRouterJun 2OpenRouter has integrated Microsoft’s new in-house MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2 models into its unified API. These models provide a high-performance stack for image, speech, and voice tasks built entirely without third-party distillation.
OpenRouterJun 2OpenRouter has added DigitalOcean’s AI-Native Cloud as an infrastructure provider for high-performance model hosting. The integration delivers industry-leading output speeds for DeepSeek V3.2, allowing developers to prioritize low-latency responses in agentic workflows.
OpenRouterJun 1OpenRouter introduced a new parameter that lets users manually balance model performance against token costs on a 0–10 scale. The update gives developers granular control over how the Auto Router selects from frontier models based on prompt complexity and budget.
OpenRouterJun 1OpenRouter integrated MiniMax-M3, an open-weight multimodal model featuring a 1-million-token context window and specialized sparse attention. By reducing long-context compute costs by 95%, the model enables persistent agentic workflows across massive codebases and video files.
OpenRouterMay 30OpenRouter launched Guardrails, a suite of security and governance tools for managing budgets, data retention, and prompt injections across its unified API. By moving these controls to the routing layer, developers can enforce enterprise-grade safety and cost policies without rewriting code for individual model providers.
OpenRouterMay 29OpenRouter introduced apply_patch, a server-side tool that enables any LLM to propose file creations, updates, or deletions using a unified diff format. By validating diff syntax server-side, the tool allows developers to build model-agnostic coding agents without writing custom parsing logic for every provider's unique output style.
OpenRouterMay 29OpenRouter released a new comparison interface that visualizes live performance metrics, pricing, and token usage trends across hundreds of LLMs. By moving beyond static benchmarks, the tool helps developers select models based on actual production data like p50 latency and reasoning token volume.
OpenRouterMay 28OpenRouter integrated Anthropic's Claude Opus 4.8, delivering significant gains in agentic coding and reasoning at the same price as the previous version. The update introduces a high-speed Fast Mode that provides 2.5x throughput for only twice the standard cost.
OpenRouterMay 26OpenRouter secured a $113 million Series B led by Alphabet's CapitalG to expand its model routing and optimization platform. The funding follows a 5x surge in weekly token volume, signaling a massive shift toward multi-model production workloads in the enterprise.
OpenRouterMay 21OpenRouter integrated Alibaba's Qwen3.7-Max, a flagship model optimized for autonomous agent loops and multi-hour task execution. The update introduces explicit prompt caching for the Qwen series, allowing developers to maintain massive context windows at a 90 percent discount on subsequent requests.
OpenRouterMay 19OpenRouter integrated Google's Gemini 3.5 Flash, a model that outperforms the previous 3.1 Pro version in coding and tool use at a lower price point. With a 1 million token context window and adjustable thinking levels, it provides a cost-effective alternative for complex autonomous workflows.
OpenRouterMay 19OpenRouter released server-side web search and fetch tools that allow any tool-calling model to autonomously browse the live web. By standardizing the tool schema, developers can swap between providers like OpenAI and Anthropic without rewriting their search implementation or domain filters.
OpenRouterMay 18OpenRouter integrated xAI's multimodal suite, enabling developers to generate photorealistic images, short video clips, and natural speech through a single API. The update allows for complex creative workflows that combine xAI's generative models with existing reasoning and coding tools on the platform.
OpenRouterMay 18OpenRouter released a suite of primitives for its Agent SDK designed to support autonomous tasks that run for hours. These tools handle the complex state management and cost controls required to move AI agents from simple chat interactions to reliable, long-running production workflows.
OpenRouterMay 15OpenRouter overhauled its Bring Your Own Key system to support multiple provider keys with tiered priority and granular usage filters. The update allows developers to stack rate limits and automate failovers between their own credentials and shared platform capacity.
OpenRouterMay 15OpenRouter integrated the Recraft V4.1 suite, introducing six specialized models for high-fidelity photorealism, editable vector graphics, and restraint-first product imagery. The update allows developers to programmatically control brand palettes and background colors while benefiting from improved short-prompt adherence.
OpenRouterMay 13OpenRouter enabled a high-speed inference tier for Anthropic's flagship model, delivering 2.5x faster throughput at a 6x price premium. This update allows developers to trade capital for speed in latency-sensitive agentic workflows without sacrificing reasoning depth.
OpenRouterMay 12OpenRouter integrated Perceptron Mk1, a vision-language model designed for spatial grounding and video understanding with structured outputs like bounding boxes and timestamps. The model introduces a reasoning toggle for complex visual tasks at a significantly lower price point than general-purpose frontier models.
OpenRouterMay 12OpenRouter transitioned its production traffic for Claude models to the newly launched Claude Platform on AWS, confirming consistent performance and uptime. This integration gives developers the full suite of native Anthropic features while keeping existing AWS billing and security infrastructure.
OpenRouterMay 11OpenRouter integrated Ring-2.6-1T, a trillion-parameter reasoning model from AntLingAGI that is free to use through May 15th. The model introduces adjustable thinking effort to balance cognitive depth with token costs, making frontier-level reasoning accessible for production agent workflows.
OpenRouterMay 9OpenRouter released Pareto Code, an experimental router that automatically sends API requests to the most affordable coding model that meets a user-defined quality threshold. By setting a minimum performance score, developers can ensure their applications use the best bang-for-the-buck models without manually tracking the latest benchmark leaders.
OpenRouterMay 8OpenRouter introduced human-in-the-loop controls to its Agent SDK, allowing developers to build agents that pause for review on sensitive actions. The update includes hooks to auto-resolve routine tasks while surfacing high-stakes decisions to a human interface. This standardizes how autonomous agents ask for help without losing conversation state.
OpenRouterMay 8OpenRouter integrated the Recraft V4 suite into its unified API, offering models specifically tuned for art-directed composition and professional design taste. The update allows developers to programmatically control brand colors and text layouts, moving AI image generation from generic stock visuals toward functional branding and marketing assets.
Frequently asked questions