ScaleDown launches Summarize, a 1M-context abstractive summarization model at $0.04 per 1M input tokens on OpenRouter

ScaleDown has released Summarize, an abstractive summarization model with a 1M-token context window, available through OpenRouter under the ID scaledown/summarize. OpenRouter lists the release date as October 13, 2026.

Key specifications

Spec	Detail
Context window	1,000,000 tokens
Input price	$0.04 per 1M tokens (20% off the $0.05 list price)
Output price	$0 per 1M tokens
Median (P50) latency	1.05 s
Throughput (P50)	92 tokens/second
Uptime (3 days)	100.00%
Parameter count	Not disclosed
Training cutoff	Not disclosed
Benchmark scores	None published

The latency, throughput and uptime figures are OpenRouter's own measurements over the first 72 hours. ScaleDown is currently the only provider routing the model.

What the model does

According to ScaleDown, Summarize condenses long text and documents into a fluent, faithful rewrite rather than extracting sentences from the source. The company lists document review, transcript digests and RAG pre-processing as target uses. In RAG pipelines, retrieved content is shortened before it reaches a downstream model.

Callers steer format, focus, language and length through the system prompt. Sampling is fixed to a configuration that ScaleDown says is tuned for faithful output, so users cannot change sampling parameters. ScaleDown has not published faithfulness or hallucination metrics, so the claim of faithful output is unverified.

Pricing

OpenRouter shows a 20% discount on the model, bringing input to $0.04 per 1M tokens from a $0.05 list price. Output tokens are listed as free. For a summarization workload, where input is large and output is short, the effective cost is dominated by input tokens. OpenRouter has not said how long the discount will last.

Companion models

ScaleDown lists two other models on OpenRouter:

• Extract: a description-driven entity extraction model. Callers describe the fields they want in plain English via provider.options.scaledown.entities. It returns matching spans with confidence scores, character offsets and surrounding context, and includes built-in OCR for scanned images and PDFs. It has a 1M context window and is priced at $0.04 input and $0 output per 1M tokens.
• Classify-1: a structured decision model served through OpenRouter's Decisions API. It returns calibrated probabilities for choice, yes/no and ordered-scale questions instead of generated text. ScaleDown says it is request-compatible with TypeSafe's Jev Decisions API. Priced at $0.04 input and $0 output per 1M tokens.

What this means

Summarize targets a narrow job, shortening text before a larger model sees it. The pricing reflects that. At $0.04 per 1M input tokens with free output, the model is priced to run on every retrieved chunk or transcript rather than selectively. The 1M context window lets it take whole document sets in one pass, which removes the chunking logic that long-document summarization usually needs.

The missing information matters more than the price. There are no published benchmarks, no parameter count and no faithfulness evaluation, and faithfulness is the main thing a summarizer is sold on. Teams considering it for compliance-sensitive document review should test it on their own data before adopting it. The fixed sampling configuration also limits tuning.

Taken together, the three models look like a bundle of small, task-specific, low-cost endpoints, each aimed at a single pre-processing step in an LLM pipeline. If the quality holds up, that is a credible alternative to prompting a general-purpose model for the same tasks.