API model profiles

Complete OpenRouter catalogue

The directory requests GET https://openrouter.ai/api/v1/models?output_modalities=all, without limit or offset, to load the full public list. OpenRouter defaults to text output when the modality query is omitted. Free variants, aliases, routers, text, images, audio, speech, transcription, video, embeddings, rerank and decision models are all retained. New output categories appear automatically in the filter.

Search matches model names and identifiers. Filter by author, output modality, pricing type, input tariff and context; sort by release date, input price or context capacity. All models are loaded together; the homepage initially shows 10 selected models from different authors, followed by batches of 30 through Show more models. The default Featured APIs first order keeps that starter selection at the top; other sorts and filters cover the complete catalogue. Counts show matches against the entire catalogue. Refresh is checked every 60 seconds while the page is open.

One profile per catalogue model

Click a model name in the directory to open /models/{provider}/{model}. Each profile uses the model's current published catalogue entry, including its identifier, name, description, prices, context capacity and any available capability metadata.

A model disappearing upstream may make its profile unavailable. A saved community registration still retains its historical snapshot; that does not guarantee the model remains callable.

Pros and cons

The assessment is generated from verifiable catalogue fields, not an invented review or an undisclosed language-model score. It compares input and output prices with the unweighted median of entries with the same output modalities and known nonnegative token/request tariffs, including free variants. Models with multiple catalogue variants count as separate entries; this is not a market-share-weighted measure.

Large published context, absence of a separate request fee and advertised upstream capabilities are described explicitly. High relative output cost, smaller context, request fees and this gateway's text-only limits are identified as trade-offs. Lower prices do not prove better quality or suitability.

No independent quality benchmark, speed measurement, uptime history or safety rating is claimed. Provider descriptions are labelled as provider-supplied rather than presented as independent evidence.

Capability distinction

An upstream model may support images or tool calls while this gateway does not expose those parameters. Profiles display both the advertised upstream capabilities and the supported Trident Protocol request subset. The gateway limit is at most 4,096 output tokens; absent upstream output metadata is labelled not published.

The directory and public /api/models route list all models. The authenticated /v1/models route, community model selectors and inference gateway expose only models accepting text input, producing exclusively text output, publishing a positive context capacity and known nonnegative prompt, completion and request tariffs. Both paid and free text models can qualify. Other models have profiles and source links but no community-use action. Unknown rates and unsupported output formats are rejected before an inference reservation or provider completion request.

Pricing labels

Missing, invalid and negative sentinel token rates are Variable / unlisted, never zero. Free text models must have known zero token/request tariffs and no other nonzero or unknown pricing fields. Zero prompt and completion rates alone do not establish that image, video or audio output is free. Full source pricing fields, including extra charges and tier overrides, remain visible on each profile. Charts plot only known token tariffs; they do not represent total multimodal cost.

Free variants still have upstream rate limits and availability conditions. Community authentication, holdings checks, funding gates and daily limits continue to apply. The gateway temporarily reserves at least one micro-dollar and refunds it when the provider reports zero cost; a missing cost report still requires reconciliation.

Cost calculator

For models supported by the text API, the calculator multiplies user-entered input and output token counts by the observed tariffs and adds any listed request fee. It is illustrative pricing arithmetic, not a quote, charge or reservation. It does not silently convert API tokens into a guaranteed redemption value for a tradable coin.

The actual gateway separately enforces request size, context, budget and current provider availability. Unlisted charges, reasoning-token billing, caching, provider changes and other conditions may alter the final bill.

Sources and freshness

Profiles show their observation time and stale status. Their source is the OpenRouter models endpoint, whose schema is documented here. The original model page is linked from each profile.

The all-model cache uses a versioned key. Older paid-text-only snapshots are not reused as a fallback. After an upstream failure, an existing complete snapshot may be displayed as stale; inference and new reference snapshots still require a recent successful catalogue.

Homepage resilience

Ten real, dated OpenRouter observations are bundled with the application in lib/featured-models.json. The homepage renders these cards immediately, before any client request or wallet connection. A saved selection is labelled as such, with its original observation time. Successful refreshes replace it with the full current catalogue; failures keep the saved cards visible instead of showing an empty search result.

/api/models is an isolated public route. It requires neither an OpenRouter secret nor ledger credentials. Database caching and history recording are optional and bounded so a storage failure cannot discard a successful OpenRouter response. If OpenRouter is unavailable, discovery may show a stale complete cache or the dated starter selection. Inference reservations and new registration price snapshots never use the bundled fallback. Those operations still require a recent successful catalogue and all existing account, holdings and funding checks.