Updated 2026-08-16 · 1502 models · 73 providers

The same model, up to 14.0× cheaper

Open-weight models are served by dozens of independent providers at wildly different prices — for byte-identical weights. This index tracks 1502 models across text, image, video, audio and 3D, with 805 of them priced, so you can see exactly where you are overpaying.

Coverage: 496 video · 439 image · 225 text · 190 multimodal · 104 audio · 48 3d. Text models are priced per 1M tokens; media models are priced in their own billing unit (per image, per second of video), which is why the two are not mixed in one league table. Sources: OpenRouter for text models, fal.ai for media. Prices are read from the seller, not estimated. Every ranking here uses the standard on-demand tier — sellers list the same model on batch and flex tiers at around half the price in exchange for a queue, and mixing those in would rank a model that answers in hours above one that answers now. OpenRouter (default route) appears as a buyable route in its own right, marked * — it resells the hosts listed beside it, so it is excluded from host counts and from the same-model price gaps below.

Most widely-used models and their cheapest credible alternative

One model per vendor, ordered by how prominent the vendor is and how many independent providers serve it — hosts only carry what customers ask for, so that is measured demand. It is not traffic or usage data, which we do not have. A matched alternative scored within 10 points of the model on our own probe suite; unverified means it is cheaper but we have not measured one or both sides, so no quality claim is made. Both prices are standard on-demand rates, so the saving is a different model and not the same model on a queue. The Kind column separates reasoning models from standard ones: reasoning models bill hidden thinking tokens at the output rate, so an identical $/1M does not mean an identical bill.

6 of these alternatives are quality-matched against our own measurements; the rest are price-only until we measure them. Cheaper is only a real saving if the alternative also has the context window and capabilities your workload needs.

Browse by what the model produces

Categories are keyed on output: a model that reads images but writes text is a Text model, because that is what you buy it for. Newest first — in this market a release date is the most reliable signal of relevance, and it is a fact we hold for every model rather than a proxy. Counts show priced / total, since not every provider publishes a rate.

Text — 302 priced of 404

All 404 Text models, ranked →

Image — 117 priced of 446

All 446 Image models, ranked →

Video — 352 priced of 496

All 496 Video models, ranked →

Music & Sound — 17 priced of 108

All 108 Music & Sound models, ranked →

3D — 17 priced of 48

All 48 3D models, ranked →

Biggest price gaps for identical weights

Cheapest vs dearest host, standard on-demand tier only, one row per provider. Same model both ends — no quality trade-off by definition, though quantisation can still differ. Batch and flex rates are left out so a gap is never manufactured by comparing a queued rate at one host against an interactive rate at another.

Cheapest models right now

Blended USD per 1M tokens at a 3:1 input:output ratio, cheapest host per model, all on the standard on-demand tier. Discounted tiers are excluded from this ranking on purpose: batch and flex run about 50% below standard in exchange for a queue, so listing them here would put a model that answers in hours above one that answers now. Each model page shows its cheaper tiers. 3 models are also held out because two first-party sources quote prices several times apart — a reseller below the host it resells — so one of the numbers is wrong and we will not rank on it.

* OpenRouter chooses a host for each request, and the figure above is what its default choice costs — not a fixed rate for this model. Selecting a specific host through the same account costs whatever that host charges, which is why this row can sit above or below the hosts listed beside it. Token rates are passed through from the underlying provider with no markup, so the per-1M price above is what that host charges directly — OpenRouter's own charge sits on funding instead: 5.5% (min $0.80) by Stripe or 5% by Coinbase to buy credits, 5% of the equivalent spend if you bring your own provider key. So budget about 5% above the rate shown unless you already hold credit. Fee schedule.

Best value: measured quality per dollar

Quality is our own measurement — text-v1 v1 — not a vendor claim or a third-party leaderboard. Every model runs the identical probe suite at temperature 0, scored by deterministic checks with no LLM judge. Score is verifiable capability (arithmetic, format compliance, retrieval from context, refusal calibration), not prose quality. Models are measured on request, so most of the catalog below is not yet scored. Quality per dollar divides the score by the standard on-demand rate. A batch or flex rate would roughly double every figure in the last column while changing nothing about the model, which is why the ranking does not use one.

How to actually cut the bill

Browse by company

Every company with more than one model in the index. A company page shows its full range, its licence mix where the weights are public, and — the fact that only shows up in aggregate — whether you can buy from the lab itself or only through a host.

86 companies have a page; 27 of them sell their own models.

All models

Every model in the index, A–Z. Use the search box above to jump to one by name, company or job.