ScaleDown Summarize vs Extract on OpenRouter: price and context
Both rows share a 1,000,000 context and 0.04 USD per million input tokens. The listing shows an output price of zero, with completion caps of 4000 and 900000.
On 2026-11-02 the OpenRouter catalog lists scaledown/summarize next to scaledown/extract. This check reads published fields only. It is not a latency test.
Both public pages state 0.04 USD per million input tokens and 0 USD per million output tokens. That is the listed price. ScaleDown Summarize versus Extract differs in job and completion cap, not in this input price.
Price and context already on the listing
Catalog prices are USD per token. The table multiplies prompt and completion by 1,000,000. 0.00000004 × 1,000,000 = 0.04. The completion string is 0, so the listed output price is 0.00.
| Field | ScaleDown Summarize | ScaleDown Extract |
|---|---|---|
| Model id | scaledown/summarize | scaledown/extract |
| Context | 1000000 | 1000000 |
| Input (USD / 1,000,000 tokens) | 0.04 | 0.04 |
| Listed output (USD / 1,000,000 tokens) | 0.00 | 0.00 |
| prompt (USD / token) | 0.00000004 | 0.00000004 |
| completion (USD / token) | 0 | 0 |
| created (UTC) | 2026-10-13 12:09:20 | 2026-10-13 12:09:11 |
| canonical_slug | scaledown/summarize-20260928 | scaledown/extract-20260928 |
| Modality | text+image+file → text | text+image+file → text |
| max_completion_tokens | 4000 | 900000 |
| hugging_face_id | empty | empty |
| Endpoint discount | 0.2 | 0.2 |
- What an output price of 0 means
- The catalog writes completion as 0, and the public page says 0 USD per million output tokens. The listing shows zero. Summarize max_completion_tokens is 4000; Extract is 900000. A listed zero is not output free of all limits.
- discount 0.2
- The endpoint price object includes this field. This note does not turn it into a second dollar figure. The catalog and the public page still show 0.04 in and a listed 0 out, per 1,000,000 tokens.
What the blurbs say, and which fields differ
Summarize
The API blurb is cut at an ellipsis: an abstractive summarizer from ScaleDown that rewrites long text instead of lifting sentences. supported_parameters lists only max_tokens. Completion cap 4000.
Extract
That blurb is also cut: description-driven entity extraction. Callers name fields in plain English through provider.options.scaledown.entities. Completion cap 900000. Input modalities are again text, image, and file.
Hugging Face has no repo for these two ids
The same day, Hugging Face search=ScaleDown returned 8 repos. Exact searches for the two API ids returned 0. The scaledown author account has one unrelated repo.
| Query | Result | Relation to these API ids |
|---|---|---|
| search=ScaleDown | 8 repos | No official summarize or extract repo |
| search=scaledown/summarize | 0 | None |
| search=scaledown/extract | 0 | None |
| author=scaledown | persona_vector_project_artifacts | 2025-09-03, not these ids |
| ScaleDown-7B GGUF | RichardErkhov, tensorblock, mradermacher | 2024 community quantizations, unrelated |
| Dawn9705/openbee-scaledown-stage1–3 | 2026-03 | Name contains scaledown; not these ids |
Read the rows in this order
- 01Match context and input price
Both rows show context 1000000 and 0.04 USD per 1,000,000 input tokens. The raw prompt price is 0.00000004 USD per token.
- 02Then read the listed output price and the cap
completion is 0 on both, so the listing shows zero. Also read max_completion_tokens: 4000 versus 900000. Do not treat a zero price as the absence of every limit.
- 03Do not fill an empty repo
hugging_face_id is empty. The 2024 ScaleDown-7B GGUF repos are not these API ids. There is no eval score, and the shared price does not say which row answers better.
What you can check is price, context, and the completion cap. The listing shows an output price of zero. That is not output free of all limits. These rows also do not yield a latency.
After the fields match, open one shared picture in the browser via the guide.
Do the two ids share context and input price?
Yes. On the 2026-11-02 list, context_length is 1000000 on both. prompt is 0.00000004 USD per token, which is 0.04 USD per 1,000,000 tokens. completion is 0, so the listed output price is 0.00 USD per 1,000,000 tokens.
Does an output price of 0 mean output has no limits?
No. The catalog writes completion as 0, and the public page says 0 USD per million output tokens. The listing shows zero. Summarize caps max_completion_tokens at 4000; Extract caps it at 900000.
Is there an official Hugging Face repo for these models?
Not in this search. ScaleDown returned 8 repos. Exact searches for the two API ids returned 0. hugging_face_id is empty. Older ScaleDown-7B GGUF repos are unrelated.
Which date is the launch date, created or the slug suffix?
created converts to 2026-10-13 (UTC): Summarize at 12:09:20, Extract at 12:09:11. The canonical_slug suffix is 20260928. The catalog does not add a launch notice. Do not read the suffix as a ship date.