Posts report Elephant Alpha on OpenRouter with 100B params and free 256K context

• hasantoxr's first report said an unannounced model called Elephant Alpha appeared on OpenRouter with no named lab, while OpenRouter's model page lists it as openrouter/elephant-alpha.
• The published specs are unusually generous for a free listing: hasantoxr's spec rundown and OpenRouter's page both describe a 100B model with a 256K context window, 32K outputs, function calling, structured output, and prompt caching.
• Usage spiked fast. According to hasantoxr's trending post, Elephant hit No. 2 on OpenRouter Trending, and OpenRouter's rankings page says those charts are based on real usage data from models accessed through the platform.
• The strongest claim around Elephant is efficiency, not raw benchmark dominance: hasantoxr's efficiency post framed it as a lower-token, lower-latency 100B model, but the listing still does not identify the underlying lab on OpenRouter's official page.

You can open the official model page, check the live usage-based rankings, and the page even stamps a release date of Apr. 13, 2026. The strange part is that the listing is concrete about capabilities and pricing, but vague about authorship, which is why hasantoxr's attribution post ends in guesses rather than a source.

The core fact pattern is simple. Elephant Alpha showed up on OpenRouter as a usable model before any clear vendor announcement surfaced in the evidence pool.

OpenRouter's official listing names the slug, marks the model as released on Apr. 13, 2026, and describes it as a 100B-parameter text model. The same page says prompts and completions may be logged by the provider and used to improve the model, which is one of the few provider-level caveats attached to the listing.

The OpenRouter page and the tweet thread line up on the main feature set:

• 100B parameters, per hasantoxr's spec rundown and OpenRouter's model page
• 256K context window, per the same post and OpenRouter's model page
• Up to 32K output tokens, per hasantoxr and OpenRouter's model page
• Function calling, structured output, and prompt caching, per hasantoxr's feature list and OpenRouter's model page
• $0 per million input and output tokens in the surfaced listing, per hasantoxr's pricing screenshot thread and OpenRouter's model page

OpenRouter's own copy positions Elephant around code completion, debugging, document processing, and lightweight agent interactions. That is a much narrower and more useful framing than the usual mystery-model hype.

The adoption signal here is OpenRouter-specific, but it is still a real one. OpenRouter's rankings page says its charts are based on real usage data from millions of users accessing models through the service.

In the evidence pool, hasantoxr's post said Elephant reached No. 2 on the Trending chart, and the linked follow-up tied that placement to token consumption rather than benchmark scores. That does not settle quality, but it does show people were routing live traffic to it almost immediately.

The early use-case map is more practical than grand. In the thread, hasantoxr's task list singled out four workloads:

1. Code completion and debugging
2. Long-document processing in a single 256K pass
3. Lightweight agent loops where token use matters
4. Structured data pipelines with clean JSON output

That lines up with OpenRouter's description, which repeatedly sells efficiency over spectacle. The interesting part is not that a 100B model exists, it is that the surfaced packaging keeps pointing back to throughput and token discipline.

The biggest missing field is still the lab name. hasantoxr's thread ran through community guesses including DeepSeek and Qwen, but attached no confirmation.

The official listing does not resolve that either. OpenRouter's model page gives the slug, specs, release date, and usage notes, but not a model card, paper, or developer identity. For now, Elephant is a live product page with a missing byline.

One of the few hands-on reports in the evidence pool came from underwoodxie96's test, who asked Elephant to simulate four domain experts and synthesize an image-generation prompt for social engagement. The post said the model returned its reasoning and final answer in 13 seconds, then compared the result against ChatGPT on the same task.

That is still a single anecdote, but it adds one fresh data point the specs page does not: at least one creative workflow test found Elephant fast enough to feel immediate, not just cheap.