{"schema":"postcutoff/model@1","as_of":"2026-11-19T22:11:00+01:00","url":"https://postcutoff.com/m/kolibri-1/","md":"https://postcutoff.com/m/kolibri-1/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"kolibri-1","name":"Kolibri-1","org":"Aleph Alpha","family":"Kolibri","released":"2026-10-03","status":"current","type":"reasoning-llm","modality_in":["text"],"modality_out":["text"],"open_weights":true,"model_license":"apache-2.0","context_window":1048576,"max_output":null,"knowledge_cutoff":"2026-06","pricing":null,"price_line":"Price not published","access":[{"provider":"Hugging Face","url":"https://huggingface.co/Aleph-Alpha/Kolibri-1"},{"provider":"Self-hosted (vLLM + aleph-alpha-inference, OpenAI-compatible API)","docs":"https://huggingface.co/Aleph-Alpha/Kolibri-1"}],"capabilities":[{"name":"German-optimised tokenizer","detail":"Bilingual EN–DE 128k tokenizer; Aleph Alpha reports 11.2% fewer German tokens than GPT-5's tokenizer (an independent test on the German Basic Law found ~15%).","first":false,"discovered":"launch","source":"https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/"},{"name":"Sparse 1M-token context","detail":"78B-total / ~3.46B-active MoE (384 experts, 6 routed) with 512-token sliding-window attention and a global layer every 5th layer; up to 1,048,576 tokens of context (262,144 recommended).","first":false,"discovered":"launch","source":"https://huggingface.co/Aleph-Alpha/Kolibri-1"},{"name":"Abstention / grounding training","detail":"Trained with a Merlin–Arthur protocol to abstain rather than hallucinate; Aleph Alpha reports an M/A grounding score of 0.23 (scale 0–0.5).","first":false,"discovered":"launch","source":"https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/"}],"entry":"2026-10-03-aleph-alpha-kolibri-open-weights","notes":"Reasoning mode with configurable effort and tool calling. Recommended sampling temperature 1.0, top_p 0.97, top_k 128. Needs ~2×80GB GPUs (min 2× A100 80GB). Benchmarks are Aleph Alpha's own (AIME 2025 96.9 EN, GPQA Diamond 84.3 EN). Weaker than Qwen models on closed-book recall and agentic coding per reviewers. No hosted API price found.","verified":"2026-10-03","body_md":"Aleph Alpha's open-weight English–German MoE. Serve with vLLM plus the `aleph-alpha-inference` package (reasoning and tool-calling parsers enabled) for an OpenAI-compatible endpoint.","page_url":"https://postcutoff.com/m/kolibri-1/","events_after":1923,"major_after":451,"historic_after":84,"missing_at_launch":664,"events_since_release":null,"you_url":"https://postcutoff.com/you/kolibri-1/","briefings":{"xs":"https://postcutoff.com/briefings/cutoff-2026-06-xs.md","s":"https://postcutoff.com/briefings/cutoff-2026-06-s.md","m":"https://postcutoff.com/briefings/cutoff-2026-06-m.md","full":"https://postcutoff.com/briefings/cutoff-2026-06.md"}}