Space Bunny Alpha returns to OpenRouter after outage

The anonymous model known as Space Bunny Alpha is back on OpenRouter after a temporary shutdown. It is a "stealth model" — developed and operated by a third-party provider that has chosen to remain anonymous during the preview phase. OpenRouter acts only as a request router; it is not the developer, owner or supplier of the model.
Technical specifications and benchmarks
The model offers a 1 million token context window, native multimodal input, strong coding capabilities according to the operator, and adjustable reasoning effort. According to figures published by OpenRouter, throughput stands at 93 tokens per second (P50, best among providers) and latency at 1.29 seconds (P50, best provider). Two scores are also listed — 79.96% and 71.47% — but the source does not state which benchmarks were used or what baselines served for comparison, so their significance cannot be assessed.
Operating model and data policy
The commercial arrangement is unusual: the anonymous provider reserves the right to retain prompts and responses but states they are not used for training. All other use is subject to OpenRouter's "Stealth Model Terms." The model is free as of the launch date, 23 September 2026, making it an accessible option for developers who want to test a large model with a wide context window without financial commitment.
Resilience and automatic routing
OpenRouter highlights automatic recovery: when an upstream provider errors, the system routes the request to another healthy provider, provided the request filters allow it. Per-provider uptime data is accessible through the Endpoints API, enabling programmatic monitoring and control.
Bottom line
Space Bunny Alpha returns as a free model with generous specifications on paper — 1 million tokens, multimodality and fast inference — but the developer's identity remains unknown and the published benchmarks are only partial. For developers seeking an available alternative for quick experiments it may suffice; for those who require full transparency on model provenance, training data and standardised test results, the picture is still incomplete.