Ready when you are
Run the same prompt through both. Keep the one that fits.
A Fire account is free and the first key takes a minute. Pay only when you top up.
Both put many providers behind one key and neither stores your prompts unless you ask. The differences are in what comes back with each call, what else the key can do, and how big the catalog is. This page says where OpenRouter is the better fit as plainly as where Fire is.
OpenRouter facts are taken from its own data-collection documentation; everything in the Fire column is verifiable in /v1/capabilities and the docs.
| Dimension | Fire | OpenRouter |
|---|---|---|
| Prompt storage by default | None. Metadata only (model, tokens, price, call_id). | None. Prompt retention is always opt-in. |
| Opt-in logging | private (your dashboard, retention you choose) or improve (lets Fire use inputs/outputs to improve Fire). No discount for sharing. | Private input/output logging, or "use of inputs/outputs" in exchange for a 1% discount on model usage. |
| Per-call withhold | X-Fire-Content-Logging: off header on any call. | Not documented in the source cited. |
| Upstream retention control | Not a routing filter today. | Can restrict routing to zero-data-retention endpoints, globally or per request. |
| Price returned with the call | Yes — meta.price.usd is the exact charged amount on every response. | Usage and cost are available; see OpenRouter's docs for how. |
| Catalog size | 53 models, 19 providers — every one live-tested on onboarding and re-audited weekly. | Far larger. The better choice for the long tail. |
| Beyond chat | Images (flat per-image price), image edit & upscale, text-to-speech, transcription, decision models — same key. | Primarily a chat-completions gateway. |
| Multi-agent orchestration | /v2/flows: DAGs of named agents, parallel branches, repeat macros, human-gate steps; queued and billed per step. | Not a gateway feature. |
| Feedback on a call | POST /v1/calls/{call_id}/report and /rate; ratings feed public rankings with sample counts. | — |
| Third-party credentials | Vault broker: Fire injects the credential server-side; the agent never holds a key. | — |
| Service discovery | Public, machine-readable GET /v1/capabilities (AIGCSEP, covenant_level 0). | Model list API. |
| Billing | Prepaid credit packs from $5; published per-model text rates; cost-plus partner tier with disclosed markup. | Prepaid credits. |
Last reviewed October 2026. If something in the OpenRouter column is out of date, tell us and we will fix it — the point of the page is to be right.
You need the widest possible model list, including niche and community-hosted models; you want to route only to zero-data-retention endpoints; or your workload is chat completions and nothing else.
You want images, speech, transcription and decisions on the same key as chat; you need the exact charged price in every response; you are building multi-step agents with human review; or you want a catalog where "listed" means "tested this week".
Both gateways discard content unless you opt in. OpenRouter's opt-in to
product use comes with a 1% discount. Fire's improve mode has no price attached: the data is used to
evaluate and route models better, is never sold or shared, and can be switched off — with a retention window you
set — from the portal at any time.
Every Fire response reports which happened: meta.content_logged is true
or false, so an agent can prove to its own user what was kept.
No. Fire logs call metadata — model, tokens, price, duration, call_id — on every call, and discards prompt and response content unless the account opts in to private or improve logging. OpenRouter likewise stores no prompts unless you opt in, per its data-collection documentation.
No. OpenRouter offers a 1% discount for letting it use your inputs and outputs. Fire's improve mode is a plain opt-in with no price change — we would rather the choice be about trust than a percentage.
OpenRouter, by a wide margin. Fire lists 53 models across 19 providers, every one of which has passed a live onboarding checklist and is re-audited weekly. If you need the long tail, OpenRouter is the larger catalog.
Not as a routing switch. OpenRouter lets you route only to endpoints with a zero-data-retention policy. Fire controls what Fire keeps (opt-in content, retention window, per-call override) but does not currently expose upstream retention as a routing filter.
Speech, transcription, decision models and queued multi-agent flows with human gates on the same key; a flat per-image price; the exact charged price returned on every call; call IDs you can report or rate; a vault broker for third-party credentials; and an AIGCSEP-style public capabilities document.
A Fire account is free and the first key takes a minute. Pay only when you top up.