Crouton Alpha Free

An automated catalog sweep recorded the first appearance of the id stealth/crouton-alpha in the OpenRouter catalog at 2026-11-03 21:01 UTC, with a listed price of $0.00/$0.00 per 1M tokens and a context window of 1,048,576 tokens. The entry was flagged as confirmed against a primary source, the OpenRouter catalog listing. The item notes the id does not map to any tracked family, which fits a newly appeared model rather than a known lineage, and it explicitly limits itself to the narrow facts of existence, price and context size. It carries no claim about benchmark quality, weight openness or vendor intent.

OrcaRouter's analysis of Crouton Alpha describes it as free to call since its appearance on 3 November 2026, with the provider field on the listing reading "Stealth." The post frames the model as a preview of something not yet publicly named, with its developer hidden behind that label. The piece adds a cost-comparison framing: filling the full 1,048,576-token window with a single prompt costs about sixteen cents on comparable flash models or seventy-nine cents at Gemini's rate, while a coding-agent task reading 200,000 tokens and writing 20,000 costs four cents on GLM-5.3-Flash and twenty-three cents on Gemini 3.8 Flash.

AI Briefing Daily published a model item on Nov 4, 2026 titled "Crouton Alpha: Free Stealth Model, 1M Context, 30-Day Logs," describing the model as a free stealth model with a one-million-token context window that runs on OpenCode and OpenRouter. The coverage treats the model's availability on both platforms as its defining deployment fact. The same entry reports that the provider does not train on user data while retaining 30 days of logs, and that the underlying developer remains anonymous. That retention detail gives developers a concrete operational condition to weigh alongside the free pricing and large context window.

Crouton Alpha is a new free "stealth" model listed on OpenRouter under the id stealth/crouton-alpha, released Nov 3, 2026, with a 1.0M token context window and zero listed input and output prices. It is positioned as a lightweight model for scalable coding, agentic and multimodal workflows, aimed at executing bounded sub-tasks, tool use and document analysis. The listing states the model is developed and operated by an unnamed third-party provider that chose to stay anonymous during the preview, and that OpenRouter only routes requests to it rather than developing or owning it. Prompts and completions may be retained by the provider but are not used for training, and the page reports roughly 35 tok/s throughput with 97.65% uptime.