OpenRouter

Starting today, you can use OpenRouter in 日本語, 한국어, 简体中文, Español, Português, Deutsch, and Français. We change the site every day, so we built an AI translation pipeline that runs daily, translates any text that changed, checks each translation against the English original, and sends back anything that fails to be translated again. Read how we built it: openrouter.ai/blog/announcem…

MAI Transcribe 2 requests are up 12× since launch on OpenRouter. 8.5M transcriptions last week, up from 0.7M in its first week in Sept. The whole speech-to-text category saw requests grew 3x over the same timeframe. More: openrouter.ai/rankings/trans…

The Extractions API is now live in alpha on OpenRouter. Use it to pull specific details out of text, images, and PDFs by describing the fields you need, like the names in an email or the total on a receipt, and get each one back as structured data. Read the guide: openrouter.ai/blog/announcem…

Inference is faster, more reliable, and cheaper on OpenRouter. Over the past month, users of GLM 5.3 Flash have seen TTFT fall 40%, throughput up 10%, and $/1M drop 17%. All thanks to our new routing algorithm that rewards providers serving great inference with more traffic.

KT is now a provider on OpenRouter, serving 7 models from its data centers in Korea. That includes 2 new to OpenRouter: Mi:dm K 3.0 Pro, KT’s own reasoning model, and Motif 3, trained for tasks in both Korean and English. Try them out: openrouter.ai/provider/kt

Installing is even easier now. Tell your agent to go to openrouter.ai/mcp. The page is readable by agents as well as humans, so your agent reads the setup and does it for you.

Coming soon: the OpenRouter MCP in your favorite plugin store (OpenAI, Anthropic and more), and no more logging in every week. It stays signed in while keeping your access safe.

Get even more visibility into your OpenRouter usage. Your agent can show your current key, its limits, and what the session has cost so far, without leaving the conversation.

Better agent ergonomics (AgEx). Tools are easier for agents to understand and pick, structured output where it matters, and the server tells Claude Code and Codex when to reach for it. Claude Code went from 5/8 tasks to 8/8 in our runs. @manufact scores it 100/100.

Video generation is even more powerful on the OpenRouter MCP. Your agent starts the job, comes back for it when it's done, and gets the result as a link, never raw bytes stuffed into context.

Less context burned. A model lookup used to dump ~1 MB (~250k tokens) into your agent's context. It's now 38 KB, paginated, 96% smaller. Transcripts carry timestamps and speakers in plain text so every client can use them.

Test all that OpenRouter offers before you write a single line of code. Ask your agent to send a message to 600+ models, call the Decisions API, or generate an image, speech or video, with your key, right in the chat. You see the real response before you build on it.

The OpenRouter MCP just got a big upgrade. Try all of what OpenRouter offers in your favorite agent, before even writing a single line of code - it's the easiest way to get started on OpenRouter.

10/ Among open-weight inference providers, Together climbed from 18th in June to first in October by tokens served. DeepSeek returned to second with V4.1 Flash, and InferenceNet entered the top ten at fourth.

9/ Decision models reached 21% of all requests by the week of October 26. The category only started in mid-September.