OpenRouter
Provides model API access and routing, including fallback options. Use it to call your chosen model through its API; use Some AI Stuff to inspect listed models, provider quotes and named results before choosing a route.
SOME AI STUFF / ABOUT
Find the right AI model—for you or your agent.
Compare models, provider prices and benchmark results. Browse the catalog or query it through the API and MCP.
From a model name to a practical choice.
Some AI Stuff brings model details, provider routes and named benchmark results into one catalog. Use it to choose models to test, check where they are available and understand which quoted rate applies.
Browse language, image, video and audio models, along with embeddings, rerankers, 3D and decision models. Open a model page for its capabilities, exact provider endpoints, scores and source links. Compare up to four models side by side, or save a model to this browser’s watchlist to review observed price and score changes.
Narrow the catalog, then compare.
- Choose a category and search. Look for a model name, developer or source ID. A developer makes the model; a provider offers an access route. They can be different companies.
- Set the requirements. Combine developer, provider and capability filters, such as tool calling, structured output or image input. For video, choose the operation you need. Active filter chips show what is selected and let you remove one requirement at a time.
- Choose what to rank. The Sort control sits above the results. Pick Newest, Name, a named benchmark or an available price unit. Select a category for relevant price sorts; context-window sorting applies to language models. Newest keeps released, listed and discovered dates labeled separately.
- Open the evidence. Select a model for its full provider quotes and benchmark configurations. Add similar models to comparison, then follow the provider or source link for the next step.
The quick filters have specific meanings. Free API access requires an eligible current zero-cost quote, with both input and output at zero for token billing. Open weights describes weight availability and can combine with other capabilities. Has pricing requires an eligible current numeric quote; Has benchmark results requires a named model benchmark result. Provider selection narrows the price and free-access checks to that provider. Trials, credits and paid-plan allowances are listed separately under Offers and Coding plans.
Keep the unit and configuration attached.
API quotes use USD and their stated billing unit: a million tokens, image, megapixel, second, compute second, minute, hour, thousand characters, request or generation. Input, output and cached-input rates stay separate. Compare prices within a compatible unit and check resolution, duration, context size and other settings on the provider route.
A price sort uses eligible current quotes in its selected unit. It is a starting point for inspecting routes, rather than a complete workload bill. Conditional rates, component charges and older quotes remain visible on model pages with their qualifiers. A missing price is unknown; a verified zero is shown as zero. Subscription plans retain their original currency, period and allowance.
Benchmark results keep their named metric, version, tested configuration and source. Choose a metric relevant to your task and read its score direction: lower word error rates are better, for example. Model-and-agent-harness results have their own agent benchmark page. Use the linked methodology to understand what was tested, then test your own workload.
Use the tools together.
Some AI Stuff connects discovery, quoted access routes and source-attributed measurements in one snapshot. Other services provide their own catalogs, pricing, evaluation and developer tools. Follow the links that fit your next step.
Hugging Face
Hosts model repositories, model cards, datasets and demos, and supports inference providers. Build a comparison here, then open a model’s linked card to inspect weights, usage details and licensing for that exact model.
Artificial Analysis
Measures models, agents and serving infrastructure. Some AI Stuff places retained named results alongside model details and access routes. Follow each result’s source for its methodology and measurement context.
Read model information from your code.
Explore the public website visually, use REST in your own integration, or connect a header-capable MCP client. Create an account and an API key to authorize machine requests.
Free accounts can list and inspect basic model metadata: names, developers, categories, capabilities, context windows, release dates and page links. Pro adds complete model records, provider prices, named model and agent results, plans, offers, news and source status for $4.99 USD per month. Usage limits apply across your account’s keys and machine requests. Compare access and limits.
MCP server URL: https://someaistuff.com/mcp
Use Authorization: Bearer <your account key>. Browser sign-in cookies do not authorize API or MCP requests. These interfaces read evidence; call the chosen model through its provider.
Check the dates behind the data.
Snapshot generated . Each route and result retains its own checked date. Generation time is separate from a model’s release date, a catalog listing date and a benchmark’s evaluation or publication date.
A failed source refresh keeps the last available records and their original dates. Partial collection means coverage is incomplete.
Browse freely. Manage access in your account.
Saved model IDs and appearance settings use this browser’s storage. Search and filter parameters are included in page requests. The website remains public; accounts manage API keys and subscription access. Read the Privacy notice for account, sign-in and billing information.
Clear saved models on the watchlist page, or remove this site’s browser data. External source links open the named provider or publisher.