Elephant Alpha: Free 100B‑Parameter Instant Model with 256K Context on OpenRouter

New Stealth Model Release

On 13 April 2026 OpenRouter announced a new "stealth" model called Elephant Alpha. The model offers 100 billion parameters, a 256 K token context window, and is currently free at $0 per million tokens. It also natively supports function calling, structured output, and prompt caching.

Instant Model Philosophy

OpenRouter describes Elephant Alpha as an "instant model" that emphasizes smart efficiency: it aims to minimise token consumption while delivering inference performance comparable to SOTA models of similar scale. It is not the strongest model, but it is the fastest and most token‑cheap.

Primary Use Cases

Code completion and debugging: fast, logically clear responses suitable for integration with IDEs or tools like VS Code/Cursor.

Large‑scale document processing: the 256 K window can hold an entire medium‑size codebase or long report, making it ideal for batch summarisation and information‑extraction tasks.

Lightweight agent tasks: rapid multi‑step calls enable frequent LLM invocations, fitting "daily‑driven" automation workflows.

Comparison with Other "Animal" Models

Elephant Alpha: 100 B parameters, 256 K context, efficiency‑first / code / document focus.

Hunter Alpha: 1 T parameters, 1 M context, heavy‑agent / complex reasoning.

Healer Alpha: parameters undisclosed, 262 K context, multimodal perception and execution.

Who Built It? Community Guesses

OpenRouter only states the model comes "from a well‑known open‑model lab". The community speculates:

Guess A – Qwen series (Alibaba Cloud): 256 K context and token‑efficiency align with Qwen products.

Guess B – Zhipu GLM series: Pony Alpha was later identified as GLM‑5, so Elephant Alpha might share the same source.

Guess C – Other open‑source labs: Mistral, DeepSeek, or similar large‑scale models could be responsible, though technical fingerprints are lacking.

Ecosystem Pattern

The "animal‑named + anonymous release + free testing" approach creates a unique ecosystem: model providers obtain massive token traffic for feedback and tuning at low cost, while developers gain free access to cutting‑edge capabilities.

Data‑Privacy Notice

During the alpha phase, all prompts and completions sent to Elephant Alpha are recorded for model improvement. Users are advised not to input personal, confidential, or sensitive information.

How to Access

Search for openrouter/elephant-alpha on OpenRouter or invoke it via Kilo Code in VS Code/CLI.

https://openrouter.ai/openrouter/elephant-alpha

Personal Experience

The author tested Elephant Alpha in the PI Agent framework and a custom agent, showing screenshots of code completion (e.g., a Tetris implementation), front‑end page generation, project analysis, basic dialogue, and riddles. The author notes that the true feel of the model requires hands‑on experimentation.

Source information: OpenRouter official page, Kilo AI blog, community discussion, up to 15 April 2026. Model identity is unconfirmed; speculation is for reference only.