Source
55 items in the feed. The texts are the sources' own descriptions — the content belongs to OpenRouter.
BlogOpenRouter
Code reviewers that only read a Git diff have to guess whether a change breaks anything, and their fixes come back as prose you apply by hand. This guide builds a Python reviewer on the OpenRouter Responses API that uploads your changed files into a Linux sandbox, runs checks to confirm each bug, and returns validated patches. A review takes one to four minutes and costs $0.015 to $0.045.
5 h ago openrouter.ai
BlogOpenRouter
An open-weight model publishes its trained weights, so anyone can run, fine-tune or serve it. A closed-weight model is only available from its maker. We used a year of OpenRouter usage data, our price catalog and benchmark scores to show what that difference changes for cost, providers and quality.
31 Oct openrouter.ai
BlogOpenRouter
An always-on AI agent wakes up on a schedule, checks something with tools, and reports or acts without you asking. This tutorial builds one in TypeScript with the OpenRouter Agent SDK that watches Hacker News, remembers what it has seen across restarts, stops any run at five cents, and waits for your approval before it posts.
31 Oct openrouter.ai
BlogOpenRouter
Use Claude Sonnet 5.5 as your default, Haiku 5.5 for high-volume and sub-agent work, and Opus 5.5 for long tasks where a mistake is costly. We compared the three on price, benchmarks, 11 coding tasks with hidden tests and a Claude Code bug fix. All three passed everything, so the difference was cost.
30 Oct openrouter.ai
BlogOpenRouter
Claude Code is free to install, and what you pay for is the model it calls. We publish the median cost of real Claude Code sessions on OpenRouter by model and session length. This post shows those costs, why the first request costs the most, what a month of use comes to, and how to lower the bill.
30 Oct openrouter.ai
BlogOpenRouter
If your app has a model read documents, emails or web pages, anyone who writes in them can try to give it instructions. We ran 25 attacks against four current models with each defense on and off. Labels cut attack emails, but only checking every tool call with Armor stopped all of them.
29 Oct openrouter.ai
BlogOpenRouter
Send a PDF to a model that reads files, with a strict JSON schema for one row, and you get typed rows back, even from scans and tables that run across pages. We tested five routes on five PDFs and found two failures that return a normal response with no error, and the checks that catch them.
29 Oct openrouter.ai
BlogOpenRouter
Our default routing used to send most traffic to the cheapest provider for a model. The new routing algorithm constantly monitors the performance of every provider endpoint to optimize the balance of cost, speed, and reliability for each new request.
28 Oct openrouter.ai
BlogOpenRouter
When Claude Code hits your plan's five-hour or weekly limit, you can keep working by pointing it at an OpenRouter API key. You pay per token for the same Claude models, with no usage window. This guide sets it up as a separate command, checks the cost, and switches back when your plan resets.
28 Oct openrouter.ai
BlogOpenRouter
A 529 overloaded_error from Claude means Anthropic's API is out of capacity for the model you asked for, not that your key or plan is wrong. This guide shows how to retry with backoff, send the same Claude model to another provider through OpenRouter, add a fallback model, and stop the error in Claude Code.
28 Oct openrouter.ai
BlogOpenRouter
OpenRouter will screen all inference requests for child sexual abuse material (CSAM) as they pass through our service. Screening happens in flight, during request processing. Content flagged as potential CSAM may be retained for human review, with reporting and preservation where required by law.
27 Oct openrouter.ai
BlogOpenRouter
Private Endpoints let Enterprise customers route their own fine-tuned models and dedicated deployments through OpenRouter, alongside the public catalog. Private endpoints are only visible within your organization, and requests go through the API your team already uses. Available by request on the Enterprise plan.
14 Oct openrouter.ai
BlogOpenRouter
Rasp gives OpenRouter's five-person sales team about 600 hours a month back, adds room for two more calls a day per rep, and runs for about $30 a day on Ori.
7 Oct openrouter.ai
BlogOpenRouter
ElevenLabs is now on OpenRouter with 9 Text to Speech and 2 Speech to Text models. Deliver speech with human-like emotion with Eleven v4, give your agent low-latency replies with v4 Turbo, and turn meetings into notes that show who said what with Scribe v2. Here's which model to pick, and one project built start to finish.
7 Oct openrouter.ai
BlogOpenRouter
A server-side code execution tool runs the model's commands in the provider's sandbox during your API request, so you don't provision or secure a container. OpenAI, Anthropic, and Google run code for their own models. Our openrouter:shell tool runs commands for any model on the Responses and Messages APIs, and our openrouter:bash tool does the same on the Messages API. This article compares the four on runtime, isolation, persistence, and cost, shows a complete request against our sandbox with the output it returned, and describes the jobs that still need a sandbox platform of your own.
5 Oct openrouter.ai
BlogOpenRouter
We now benchmark model routers side by side across six suites. Our Router Index weighs quality, speed, and cost so you can decide if you need a router and pick the best one for your needs.
2 Oct openrouter.ai
BlogOpenRouter
Cheap-first routing sends routine support questions to a small, inexpensive model and escalates selected hard or uncertain requests to a stronger one. This guide compares the three application-side routing patterns, separates what your application owns from what OpenRouter's Auto Router and model fallbacks handle, walks through a support flow with a copyable request, and sets out how to measure escalation rate and cost per resolved ticket.
2 Oct openrouter.ai
BlogOpenRouter
Multi-model orchestration is three layers. Workflow orchestration is planning, state, memory, and delegation, and LangGraph and CrewAI are built for it. Model routing is choosing a model per call and falling back when it fails, and provider routing is choosing which provider serves that model. OpenRouter does the second two. This article separates the layers, shows the same two-step pipeline in direct OpenRouter calls and in LangChain, and describes when to add a framework, when to use our Agent SDK, and how to put OpenRouter underneath LangGraph or CrewAI.
2 Oct openrouter.ai
BlogOpenRouter
OpenAI, Anthropic, and Google each use a different request and response shape for the same tool. Agent frameworks handle that difference in different places. Some translate one definition into each provider's format, some are native to a single provider, and some hand the question to a connector underneath. This article compares six frameworks on schema definition, translation, and MCP support, then shows how OpenRouter normalizes tool calling at the API layer so a model swap is a change to one string.
2 Oct openrouter.ai
BlogOpenRouter
A one-line prompt change can ship an agent that tells customers the wrong refund window, and nothing in a normal CI pipeline checks what the model says. This guide builds a fixed eval set for a support agent, a script that exits non-zero below a measured threshold, and a GitHub Actions job that blocks the merge when the pass rate drops.
1 Oct openrouter.ai
BlogOpenRouter
The cheapest model that clears your quality bar is usually not the model at the top of a leaderboard. This framework sets the bar for one task, measures cost per quality point across a cheap, a mid-tier, and a frontier model on your own examples, and picks the cheapest one that clears the bar with margin. It uses live catalog prices and the cost each OpenRouter response reports.
1 Oct openrouter.ai
BlogOpenRouter
Confidence-based escalation keeps most requests on a cheap model and sends only the ones it is unsure about to a stronger one. This guide covers forcing a numeric confidence field with structured outputs, setting the threshold from error rates on your own traffic, tuning it against accuracy, cost, and latency, routing the escalation in your code, and re-tuning it after launch.
1 Oct openrouter.ai
BlogOpenRouter
An agent can call the wrong tool, or call the right tool with the wrong arguments. This guide covers three ways to test each failure mode, a Python harness that grades both, and how to run the same test cases against several tool-capable models through OpenRouter.
30 Sept openrouter.ai
BlogOpenRouter
A golden eval dataset is a curated set of production inputs with reviewed expected outputs, versioned in Git and run before every deploy. This guide covers the five steps to build one from live traffic and how to run the same set against many candidate models through one API.
30 Sept openrouter.ai
BlogOpenRouter
An agent's behavior can change when you edit a prompt, swap a model, change a tool schema, or change what retrieval returns. This guide covers the locked case set, the per-case behavioral contract, and how to run the same suite against two concrete model slugs through OpenRouter so the diff shows what the change did.
30 Sept openrouter.ai
BlogOpenRouter
If you already have the image a video should start from, the model choice comes down to what has to happen after that frame. This post compares the Veo 3.1, Seedance, Kling, and Grok Imagine Video lines on duration, resolution, first-frame and last-frame control, generated audio, reference inputs, and current per-second price, then shows how to submit an image-to-video job with the TypeScript SDK.
29 Sept openrouter.ai
BlogOpenRouter
We audited our own OpenRouter account and found over 1,000 active API keys for 85 employees, 168 of them hadn't been used in months. The new Security Center helps you manage and de-risk your API keys with one place to see every key, know which ones are risky and why, clean them up, and lock them to your own network.
28 Sept openrouter.ai
BlogOpenRouter
Qwen 3.8 is Alibaba's August 2026 model generation. Four models share the name and they differ in whether you can download them, which license applies, whether they read images and video, and whether you can turn reasoning off. This post covers each model, the license terms, local serving, benchmark scores, and what changes your bill on OpenRouter.
28 Sept openrouter.ai
BlogOpenRouter
ByteDance's Seedance video models are available through hosted APIs, and we found no downloadable checkpoint or model license for any of them. This post covers where we looked, what the repositories named Seedance on GitHub and Hugging Face actually contain, when a checkpoint is a hard requirement, and how to run a Seedance job through the asynchronous video endpoint on OpenRouter.
25 Sept openrouter.ai
BlogOpenRouter
Kimi K3 ships public weights under Moonshot AI's own Kimi K3 License, which is not an OSI-approved open-source license. This post explains the difference, what the license grants and requires, what the checkpoint contains, and how to call the model on OpenRouter with reasoning, vision, and tool calling.
24 Sept openrouter.ai
BlogOpenRouter
A method for putting Jev to work on a new problem, worked end to end on marketplace listing moderation: what stays in code, what Jev sees, how to phrase the questions, and how to turn its probabilities into publish, hold, or reject.
23 Sept openrouter.ai
BlogOpenRouter
An embedding model decides what your retrieval system can find. We shortlisted the embedding models in our catalog for English RAG, multilingual retrieval, code search, text-and-image retrieval, and low-cost indexing, sent live requests to each one, and recorded their prices, context windows, and default vector sizes so you can pick a candidate and test it on your own data.
23 Sept openrouter.ai
BlogOpenRouter
Claude Opus 5 leads OpenRouter's classification task ranking by spend. We sent the same 3,080 Banking77 utterances to it and to Jev 1.13 through the Decisions API. Opus scored 84.4% to Jev's 81.0%, and Jev answered in 175 ms at $0.11 per thousand requests against 2.3 seconds and $2.42 for Opus.
22 Sept openrouter.ai
BlogOpenRouter
Send a whole workload in one POST, collect the results within 24 hours, and typically pay half the per-token price. Across 230k+ batches that completed over our two week beta period, the median finished in 7 minutes.
22 Sept openrouter.ai
BlogOpenRouter
Nemotron 3.5 Lightning is NVIDIA's open-weight 30B mixture-of-experts model with about 3B active parameters per token, built for the high-volume execution calls in an agent run. This post covers what the architecture means, how the model compares with Nemotron 3 Ultra, what context and features each OpenRouter endpoint exposes, and how to call it with structured outputs.
22 Sept openrouter.ai
BlogOpenRouter
Jev reads text and returns typed answers with probabilities instead of prose. Here's what that means, three live API calls, how to read the numbers, and how to call it with an OpenRouter key.
21 Sept openrouter.ai
BlogOpenRouter
An LLM judge writes a verdict. Jev returns a probability. We graded the same labeled answers and expert-rated summaries with both, and the difference decides which one you should use for a given rubric. Jev matched the LLM judge on agreement at a fifth of the cost and a tenth of the latency, and its probabilities meant what they said. The LLM judge won on the open rubric.
21 Sept openrouter.ai
BlogOpenRouter
Descript, an AI tool for video and audio editing, solved a slow problem in testing new AI models. Previously, testing a new model took a few hours of work plus a week of waiting, because an engineer had to manually add it. By using OpenRouter and automation with Claude Tag (Anthropic's Slack integration), testing now takes only one to two hours, and anyone can start an evaluation directly from Slack. This means Descript now tests models multiple times a week instead of choosing a model early and sticking with it. The team can now quickly integrate new models — for example, adding Grok 4.5 from xAI without building its own integration.
21 Sept openrouter.ai
BlogOpenRouter
Jev returns typed judgments with probabilities. An LLM returns prose. We benchmarked both on 100 support cases through OpenRouter, then wired them together so Jev routes and the LLM writes.
19 Sept openrouter.ai
BlogOpenRouter
OpenRouter tested 20 image generation models through one API using the same prompt to compare pricing and features. Models charge in different ways—per megapixel, per token, or per image—so listed prices cannot be directly compared. By reading actual billing costs from API responses, they showed that the same image cost between $0.006 and $0.134 depending on the model, a 22-fold difference. OpenAI's gpt-image-2 was cheapest, while settings like image quality on OpenAI models could double the price. Not all models supported the same features: some could generate multiple images per request, some had no random seed support, and they returned different file formats.
18 Sept openrouter.ai
BlogOpenRouter
A tool-calling agent loop sends a conversation and tool definitions to an AI model, executes the tool calls the model requests, appends the results, and repeats until the model responds or a stop condition fires. The guide shows how to build this loop in TypeScript with the OpenRouter SDK and adds safety mechanisms: an iteration cap, detection of repeated calls, fallback to alternative models, and conversation history controls. Your application decides when to stop the loop and executes the tools the model requests.
17 Sept openrouter.ai
BlogOpenRouter
DeepSeek V4 is not a single model but a family with several versions. Of these, only two accept images: V4.1 Flash and V4 Flash Vision Exp (marked experimental). The other versions—V4 Pro 0813, V4 Flash 0731, and older checkpoints—only accept text. To use a text-only model with images, you can run a separate vision model first (such as Qwen3.8 27B or Kimi K3) and send its text description to the V4 model.
16 Sept openrouter.ai
BlogOpenRouter
Descript, an AI-powered video editing platform, previously maintained three separate integrations to different AI models. Testing and deploying a new model took over a week, largely because engineers were busy and each model had its own technical requirements. With OpenRouter as an intermediary, Descript now runs 13 models from Anthropic, Google, OpenAI, and xAI simultaneously. A new model can be tested and deployed in one to two hours, often several times a week—speed enabled by both simpler technical integration and automated testing through Slack.
15 Sept openrouter.ai
BlogOpenRouter
LLM-as-a-judge is a technique where a second AI model automatically scores another model's output against criteria you write in plain language. It fills a gap: an agent can pass all technical tests but still give a poor answer—for example, omitting a critical detail. The guide shows how to use LLM-as-a-judge with Ori Eval, calibrate thresholds against human-labeled examples, and control costs. It works best when the requirement is clear but not a single correct answer—such as verifying that an answer is grounded in actual data or follows a specific tone.
14 Sept openrouter.ai
BlogOpenRouter
Zero Data Retention (ZDR) means an AI provider processes your query and returns an answer without storing either afterward. It is a storage guarantee, but it only covers the provider's storage — not what your own systems log, not data in transit, and not third-party tools connected to the request. On OpenRouter, you can enforce ZDR at the account level, through guardrails, or with a special field in each API call. ZDR differs from "no training" (the company does not train on your data) and from region pinning (where the request is processed) — these are separate controls.
11 Sept openrouter.ai
BlogOpenRouter
OpenRouter offers a text-to-speech service through an API that works consistently regardless of which provider you use — Mistral, xAI, Microsoft, and others. You send text, select a model and voice, and receive an audio file in return. The guide shows how to authenticate with an API key, make your first request using cURL or Python, verify that the response contains audio and not just error messages, and switch between different voices and providers by changing only a few lines of code.
11 Sept openrouter.ai
BlogOpenRouter
OpenRouter Presets is a way to save all settings for an AI model in one place — which model you use, what instructions it should receive, and technical parameters like temperature. Instead of entering the same information throughout your code, you can give these settings a name and reference it. When you change the settings on OpenRouter's dashboard, all your apps automatically update without needing to rewrite code.
10 Sept openrouter.ai
BlogOpenRouter
OpenRouter Fusion is a system that sends a prompt to multiple AI models simultaneously, lets them respond in parallel, and a judge compares the responses before a final model writes its answer. It costs approximately four to five times more and takes two to three times longer than a regular call, but can provide better answers for complex questions. Fusion works best for research and analysis where multiple perspectives improve the response, but is not suitable for quick chat conversations or tasks that need to return exactly the same answer every time.
10 Sept openrouter.ai
BlogOpenRouter
OpenRouter launches US In-Region Routing, a way to guarantee that your data is handled entirely within the USA or EU. You send your requests to us.openrouter.ai or eu.openrouter.ai instead of the regular address, and everything is decrypted and processed within that region—even for models from Chinese companies like DeepSeek and GLM, as long as a US or European provider hosts them. If no local provider can serve the model you need, the request fails rather than being sent outside the region. The feature is available on Business and Enterprise plans.
9 Sept openrouter.ai
BlogOpenRouter
Seedance 2.5 is ByteDance's video model that can generate up to 30 seconds of video in 480p or 720p resolution. It costs approximately $0.10 per second in 480p and $0.23 per second in 720p. The model excels at capturing long shots and editing or extending existing video — when you input an existing video clip, it costs roughly 40 percent less than generating from scratch. Compared to the earlier version Seedance 2.0, 2.5 can produce longer video clips (30 seconds instead of 15), but it only reaches up to 720p instead of 4K resolution.
9 Sept openrouter.ai
BlogOpenRouter
OpenRouter has enabled image editing with text instructions through its API. You send a source image and a text description of desired changes in a single request and receive the edited image back. The guide demonstrates how to do this in Python and TypeScript, and explains that you can easily switch to a different editing model by changing a field in the code. You can also make multiple edits in sequence by sending each result back as a new source image.
9 Sept openrouter.ai
BlogOpenRouter
OpenRouter launches two new features: a shell tool that gives AI models access to an isolated Linux environment where they can run commands, and a Files API that allows models to upload files to work with and download results. Together, these enable models to execute entire workflows—such as web searches, script writing, and execution—completely on the server. The features are available from day one in beta version, costing $0.0001 per second for the sandbox.
8 Sept openrouter.ai
BlogOpenRouter
OpenRouter offers a unified API for video generation that abstracts the complexity of different video providers. Instead of integrating each provider separately — Seedance, Veo, Wan, and others — you use the same endpoint, authentication, and job queue logic for all. You send a prompt to POST /api/v1/videos, receive a job ID, poll until the video is ready, then retrieve the MP4. Because video generation takes several seconds to minutes, OpenRouter uses an asynchronous system where your app can continue while generation runs in the background, and the job is stored so you can retrieve it later even after restart.
25 Aug openrouter.ai
BlogOpenRouter
OpenAI offered substantial discounts on its new Terra and Luna models between July 27 and August 14. During this period, Terra token usage increased 5.6 times and Luna token usage 13.8 times, while the non-discounted Sol model increased only 1.1 times. Approximately one-third of users who tried the discounted models continued using them after the discount period ended. Most gains came from competitors rather than from other OpenAI models — OpenAI's total market share grew from 7.1 percent to 12.4 percent during the discount period.
25 Aug openrouter.ai
BlogOpenRouter
There is no universally best AI model — only the best one for your specific task, budget, and time. OpenRouter presents a six-step framework for choosing the right model: first define your task concretely, create a shortlist from benchmarks and live data, compare price and response time across providers, test the best candidates with your own instructions, evaluate them by cost per completed task (not per token), and expect the answer to change when new models launch. Through OpenRouter's MCP server, you can do all this directly in your code editor without leaving it.
25 Aug openrouter.ai
