The newest closed models charge anywhere from 12 cents to $150 per million output tokens — a 1,250-fold spread

Token pricing for the 88 most recently added private-weight (closed) model records in the OpenRouter API catalog, observed 2026-08-28. Prices are in US dollars per token and per million tokens, spanning catalog additions from 2023-05-28 to 2026-08-11 across 18 provider slugs.

88 records, not 100. The request asked for the last 100 private-weight models, but only 88 currently listed records in the live catalog passed the reproducible private/closed-weight proxy and had positive, nonzero token pricing — so this report covers 88.

88qualifying API records

18provider slugs

$1.00 / $4.34median input / output, per million tokens

$0.03 → $150lowest input to highest output, per million tokens

Every record, input to output price, newest first

Each row is one billable API record; the bar stretches from its input price to its output price, on a log scale. Hover a row for exact per-token figures. Batch and fast variants are separate billable records and stay separate.

input price ($/M tokens) output price ($/M tokens) dashed lines mark the medians

Output tokens cost more than input tokens on every one of the 88 records; at the median the markup is over four-fold ($1.00 in vs $4.3375 out per million). openrouter.ai

The cheapest inputs are Upstage Solar Pro 4 and Qwen3.7 Flash at $0.03 per million — one thousandth the price of Claude Opus 4.7 (Fast) and GPT-4 at $30 per million input. openrouter.ai

Claude Opus 4.7 (Fast) tops the output scale at $150 per million — 1,250× the cheapest output, Solar Pro 4 at $0.12. openrouter.ai

Batch variants are consistently half the interactive price where paired (Claude Opus and Sonnet lines, Gemini Flash lines, GPT-4o, GPT-4 Turbo), while the four GPT-5.6 Luna and Terra batch records list at the same price as their standard counterparts. openrouter.ai

All 88 records

Click a column header to sort. Per-token prices are exact catalog values in scientific notation; per-million columns are the same prices scaled for readability. Cached-input is blank where the catalog does not state it.

#	Provider	Model	Model ID	Added	In $/tok	Out $/tok	Cached $/tok	In $/M	Out $/M	Cached $/M	Context

Method: OpenRouter API catalog (openrouter.ai/api/v1/models) pricing observed on 2026-08-28 — 88 rows, one per currently listed billable API record. "Private-weight" is a catalog proxy: no Hugging Face model ID and no open-weight/open-source wording in the description; it is not a legal determination of license status. "Last" means newest by OpenRouter's created timestamp, not the developer's release date. Free listings, routing/meta-products, aliases beginning with ~, and records without positive token pricing were excluded — which is why 88, not 100, records qualify. Prices are USD per token as billed through OpenRouter and may differ from direct-provider pricing or change later. Batch and fast variants are separate billable records and are kept separate.