Open-Weight API Prices: 7.4× Gap at the Frontier

Frontier open-weight models cost a median of $0.544 per million tokens, compared with $4.00 for proprietary models. The proprietary median is 7.4× the open-weight median—the largest proportional price gap across the capability tiers we compared.

The answer is yes at the capability-tier level: open-weight models have lower median API prices in every tier. The size of the advantage varies, however, so a frontier budget and a small-model budget have different price references.

Capability here uses the Artificial Analysis intelligence index, a single benchmark score that does not capture every task and is not available for every model. Prices are OpenRouter list prices; providers’ own API prices, volume discounts and regional pricing can differ—see our methodology.

The mid-tier has the narrowest price gap

Among mid-tier models, the open-weight median is $0.637 per million tokens, versus $2.03 for proprietary models. The proprietary median is 3.2× the open-weight median.

This tier has the smallest proportional gap. For teams budgeting for mid-tier capability, the open-weight price advantage remains, but the frontier ratio would overstate it. Use the mid-tier medians when comparing budgets at this capability level.

Small open-weight models have the lowest median price

Small open-weight models cost a median of $0.305 per million tokens. The proprietary median in that tier is $1.20, or 3.9× the open-weight median.

The small-model tier has the lowest median price for both open-weight and proprietary models. Its proportional price gap falls between the mid-tier and frontier gaps.

For a team whose capability target falls in the small-model tier, the open-weight median is therefore the lowest price reference in this comparison. The mid-tier open-weight median is actually higher than the frontier open-weight median; the reported prices do not rise in step with capability tiers.

Provider availability is equal at the median

Open-weight models are served by a median of 3.0 providers, against 3.0 for proprietary models. Buyers can shop around in either category.

The lower open-weight median prices come alongside equal median provider counts, rather than a higher count for open-weight models. For procurement, that means the provider-count comparison offers no median availability advantage to either category.

MiMo-V2.6-Pro leads open-weight capability

MiMo-V2.6-Pro is the most capable open-weight model in our data, with an intelligence index of 46.3. Its price is $0.544 per million tokens, matching the frontier open-weight median.

That gives buyers a specific model to review at the top of the open-weight capability ranking. We track MiMo-V2.6-Pro prices alongside the broader open-weight model rankings.

• For frontier budgets, include open-weight options. The frontier comparison has the largest proportional median price advantage.
• For mid-tier budgets, use that tier’s prices. Its proprietary-to-open-weight ratio is the narrowest in the comparison.
• For small-model budgets, start with the small-tier median. It is the lowest open-weight median reported here.
• For the highest-capability open-weight option, review MiMo-V2.6-Pro. It leads the open-weight intelligence ranking at a price matching the frontier open-weight median.

How we measured this

Models are grouped into tiers by Artificial Analysis intelligence index (40 and above, 20 to 40, under 20) and split by whether their weights are published.

Where we compare models on one price, we use a blended price per 1 million tokens: three input tokens for every output token, the usual shape of chat traffic.

Limitations.

• Capability is measured with the Artificial Analysis intelligence index, a single benchmark score; it does not capture every task, and not every model has one.
• Prices are list prices on OpenRouter; a provider’s own API, volume discounts and regional pricing can differ.

Data: LLM Cost Index snapshot of 2026-10-21. Full details in our methodology.