Aleph Alpha Kolibri pricing and availability

Checked October 3, 2026 against Aleph Alpha's official launch and model card.

No official Kolibri API token price is published. Kolibri-1's weights are downloadable under Apache 2.0. That license has no per-token model royalty, but inference is not free: hardware or cloud rental, power, memory, operations, and support still cost money. We do not insert an invented $0 row into the priced API table or JSON feed.

What can you buy today?

Aleph Alpha has released Kolibri-1 weights and its model card and documents a self-hosted vLLM route. Its launch announcement and model page do not quote a public Kolibri input or output token rate. Ask Aleph Alpha for a current enterprise offer if you need managed service; do not apply historical Aleph Alpha rates to this new model.

Self-hosting economics

Kolibri has 78.1 billion total parameters but activates about 3.46 billion per token. The full checkpoint must still fit in memory. Its model card recommends a native 262,144-token context for latency-sensitive use and says serving was validated up to 1,048,576 tokens. Real cost per million tokens depends on GPU configuration, precision, traffic, context length, utilization, and operator time. Estimate total monthly infrastructure and operations spend divided by successfully served monthly tokens, multiplied by one million; measure prompt and output volumes separately.

Read our Kolibri launch and cost analysis, then compare priced alternatives in the API table. For a managed open-model alternative, see Novita's listed models; do not assume it hosts Kolibri. We may earn a commission from that sponsored link.

Labs status

Kolibri is not in our maintained Labs roster. The published local weights are not a funded, callable exact route in our benchmark transport. A fair run needs a pinned checkpoint and serving plugin, GPU and precision settings, the same 49 tasks, full responses, latency, failures, token counts, and infrastructure-cost allocation. See the coverage note.