Kolibri-1 Benchmarks, Specifications & Availability

Explore Kolibri-1 from Aleph Alpha: published specifications, source-linked vendor benchmarks, independent evaluator coverage and recorded pricing when available.

Compare Kolibri-1 with other models →Explore data coverage

Published specifications

Provider: Aleph Alpha
Access: Open weights
License: Apache 2.0
Context window: 1.05M
Total parameters: 78B
Active parameters: 3.46B
Released: 2026-10-03
Modalities: text
Family: Kolibri

Model card · Announcement · Website

Model notes

Official ungated FP8 weights on Hugging Face; Apache 2.0. Card states 78,103,074,560 total and 3,457,573,120 active parameters. English/German reasoning and tool calling. Native context 262,144 tokens; validated extrapolation to 1,048,576 tokens; vendor recommends <=262,144 for serving efficiency and complex tasks. Benchmarks at high reasoning effort in Aleph Alpha eval-framework/Harbor. No OpenRouter listing on 2026-10-03. BF16 is a precision variant, not a separate model.

Pricing · OpenRouter

No OpenRouter pricing is recorded for this model. Missing rates are not free.

Official / vendor benchmarks

Default headline records. Own-vendor, peer-vendor and third-party provenance remain visible in evidence; configurations may differ.

Official / vendor headline scores; expand evidence for every record
Benchmark / evaluator	Headline score	Evidence
Aleph Alpha post-training Overall (EN)	75.5%high effort	All 1 recorded result & sources 75.5% · raw 75.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha post-training Overall (DE)	70.8%high effort	All 1 recorded result & sources 70.8% · raw 70.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Knowledge average (EN)	50.1%high effort	All 1 recorded result & sources 50.1% · raw 50.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Knowledge average (DE)	57.6%high effort	All 1 recorded result & sources 57.6% · raw 57.6 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
GPQA Diamond	84.3%high effort	All 1 recorded result & sources 84.3% · raw 84.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Chain-of-thought, avg@8. https://huggingface.co/Aleph-Alpha/Kolibri-1
GPQA Diamond (DE)	81.3%high effort	All 1 recorded result & sources 81.3% · raw 81.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. https://huggingface.co/Aleph-Alpha/Kolibri-1
HLE (text, no tools)	21.5%high effort	All 1 recorded result & sources 21.5% · raw 21.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Text-only questions, no output cap. https://huggingface.co/Aleph-Alpha/Kolibri-1
Humanity's Last Exam (DE)	15.9%high effort	All 1 recorded result & sources 15.9% · raw 15.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. https://huggingface.co/Aleph-Alpha/Kolibri-1
AA-Omniscience Accuracy (public set)	14.8%high effort	All 1 recorded result & sources 14.8% · raw 14.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
AA-Omniscience Index (public set)	-32.8high effort	All 1 recorded result & sources -32.8 · raw -32.8 index Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
MMLU-Pro CoT (EN)	80%high effort	All 1 recorded result & sources 80% · raw 80 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
MMLU-ProX CoT (DE)	75.5%high effort	All 1 recorded result & sources 75.5% · raw 75.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Math average (EN)	96.5%high effort	All 1 recorded result & sources 96.5% · raw 96.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Math average (DE)	88.8%high effort	All 1 recorded result & sources 88.8% · raw 88.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
AIME 2025 (EN)	96.9%high effort	All 1 recorded result & sources 96.9% · raw 96.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). avg@16. https://huggingface.co/Aleph-Alpha/Kolibri-1
AIME 2025 (DE)	87.5%high effort	All 1 recorded result & sources 87.5% · raw 87.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. avg@16. https://huggingface.co/Aleph-Alpha/Kolibri-1
AIME 2026 (EN)	96%high effort	All 1 recorded result & sources 96% · raw 96 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). avg@16. https://huggingface.co/Aleph-Alpha/Kolibri-1
AIME 2026 (DE)	90%high effort	All 1 recorded result & sources 90% · raw 90 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. avg@16. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Agentic average (EN)	63.4%high effort	All 1 recorded result & sources 63.4% · raw 63.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Terminal-Bench 2.1	27.7%high effort	All 1 recorded result & sources 27.7% · raw 27.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Terminus 2 agent; max 250 steps; reply cap 80k tokens or half the context; avg@3. https://huggingface.co/Aleph-Alpha/Kolibri-1
Tau2-Bench (Telecom)	94.7%high effort	All 1 recorded result & sources 94.7% · raw 94.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Domain score, avg@3; not the three-domain aggregate. https://huggingface.co/Aleph-Alpha/Kolibri-1
Tau2-Bench (Retail)	69.9%high effort	All 1 recorded result & sources 69.9% · raw 69.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Domain score, avg@3; not the three-domain aggregate. https://huggingface.co/Aleph-Alpha/Kolibri-1
Tau2-Bench (Airline)	76.7%high effort	All 1 recorded result & sources 76.7% · raw 76.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Domain score, avg@3; not the three-domain aggregate. https://huggingface.co/Aleph-Alpha/Kolibri-1
τ³-Bench Banking	38.1%high effort	All 1 recorded result & sources 38.1% · raw 38.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Banking alltools retrieval configuration; avg@4; context overflow scores zero. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v3 (multi-turn)	39.8%high effort	All 1 recorded result & sources 39.8% · raw 39.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (overall)	61.4%high effort	All 1 recorded result & sources 61.4% · raw 61.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (non-live AST)	79.1%high effort	All 1 recorded result & sources 79.1% · raw 79.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (live)	78.9%high effort	All 1 recorded result & sources 78.9% · raw 78.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (multi-turn)	47.5%high effort	All 1 recorded result & sources 47.5% · raw 47.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (memory)	62.8%high effort	All 1 recorded result & sources 62.8% · raw 62.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BFCL v4 (web search)	62.5%high effort	All 1 recorded result & sources 62.5% · raw 62.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). bfcl-eval; different web-search backend and per-model sampling; not comparable to BFCL leaderboard. Overall includes irrelevance detection not separately printed. https://huggingface.co/Aleph-Alpha/Kolibri-1
BrowseComp	29.4%high effort	All 1 recorded result & sources 29.4% · raw 29.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Code average (EN)	89.3%high effort	All 1 recorded result & sources 89.3% · raw 89.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
LiveCodeBench v6	85.9%high effort	All 1 recorded result & sources 85.9% · raw 85.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). August 2024-April 2025 problems; avg@3. https://huggingface.co/Aleph-Alpha/Kolibri-1
HumanEval+	92.7%high effort	All 1 recorded result & sources 92.7% · raw 92.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SWE-bench Verified	66.4%high effort	All 1 recorded result & sources 66.4% · raw 66.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). OpenCode agent; at most 250 steps; dedicated deployment. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Instruction Following average (EN)	78.1%high effort	All 1 recorded result & sources 78.1% · raw 78.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
IFBench (loose-prompt)	78.1%high effort	All 1 recorded result & sources 78.1% · raw 78.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Loose-prompt metric, avg@5; distinct from strict IFBench scoring. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Grounding / Hallucinations average (EN)	59.4%high effort	All 1 recorded result & sources 59.4% · raw 59.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
SQuAD (M/A Grounding Score)	23.4%high effort	All 1 recorded result & sources 23.4% · raw 23.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SQuAD (Utility Accuracy)	83.4%high effort	All 1 recorded result & sources 83.4% · raw 83.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
RGB Closed-Book	51%high effort	All 1 recorded result & sources 51% · raw 51 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
RGB Negative (Abstention)	85.6%high effort	All 1 recorded result & sources 85.6% · raw 85.6 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
AA-Omniscience Non-Hallucination Rate (1 − Hallucination Rate, public set)	44%high effort	All 1 recorded result & sources 44% · raw 44 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
RGB Fact-Check (Error Correction)	34%high effort	All 1 recorded result & sources 34% · raw 34 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
FRAMES (<24k)	71.2%high effort	All 1 recorded result & sources 71.2% · raw 71.2 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
FRAMES (>24k)	73%high effort	All 1 recorded result & sources 73% · raw 73 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SealQA (no distractors, <24k)	80.7%high effort	All 1 recorded result & sources 80.7% · raw 80.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SealQA (12 distractors, <24k)	61%high effort	All 1 recorded result & sources 61% · raw 61 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SealQA (no distractors, >24k)	100%high effort	All 1 recorded result & sources 100% · raw 100 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
SealQA (12 distractors, >24k)	65.1%high effort	All 1 recorded result & sources 65.1% · raw 65.1 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Agentic Retrieval average (EN)	77.3%high effort	All 1 recorded result & sources 77.3% · raw 77.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Agentic Retrieval average (DE)	69.4%high effort	All 1 recorded result & sources 69.4% · raw 69.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
MuSiQue (EN)	77.3%high effort	All 1 recorded result & sources 77.3% · raw 77.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Honeypot	80.8%high effort	All 1 recorded result & sources 80.8% · raw 80.8 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Agentic Wiki QA (DE)	69.4%high effort	All 1 recorded result & sources 69.4% · raw 69.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Industry RAG average (EN)	89.7%high effort	All 1 recorded result & sources 89.7% · raw 89.7 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Aleph Alpha Industry RAG average (DE)	67.5%high effort	All 1 recorded result & sources 67.5% · raw 67.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). German translation (ellamind), except MMLU-ProX and vendor retrieval/proxy tasks. Vendor aggregate excludes incomplete/non-percentage rows; Overall EN uses eight categories, DE four. Long Context excluded. https://huggingface.co/Aleph-Alpha/Kolibri-1
Semiconductors	80.4%high effort	All 1 recorded result & sources 80.4% · raw 80.4 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
German Public Sector	75%high effort	All 1 recorded result & sources 75% · raw 75 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Aerospace	58.9%high effort	All 1 recorded result & sources 58.9% · raw 58.9 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Automotive Supplier	99%high effort	All 1 recorded result & sources 99% · raw 99 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
Industrial Drive Technology	60%high effort	All 1 recorded result & sources 60% · raw 60 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
LongBench Pro	64.5%high effort	All 1 recorded result & sources 64.5% · raw 64.5 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). https://huggingface.co/Aleph-Alpha/Kolibri-1
AA-LCR	68.3%high effort	All 1 recorded result & sources 68.3% · raw 68.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha eval-framework/Harbor, released post-trained checkpoint. Kolibri reasoning effort high; peers use vendor-documented sampling. Technical report Tables 27-29 (pp. 96-101). Vendor-run AA-LCR, avg@3; not an independent AA measurement. https://huggingface.co/Aleph-Alpha/Kolibri-1
RGB: invents nothing	87.3%high effort	All 1 recorded result & sources 87.3% · raw 87.3 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha grounding evaluation. Release radar visually inspected; exact value from its Show the numbers companion table. Not the RGB Negative (Abstention) metric. https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/
AA-Omniscience Index (scaled public set)	33.6%high effort	All 1 recorded result & sources 33.6% · raw 33.6 % Headline · high effort · Own vendor Source/record date: 2026-10-03 Harness: Aleph Alpha public-set evaluation. Published product-page radar companion table gives 33.6% on the scaled display; the raw index is -32.8, preserved separately. Not an independent AA score. https://aleph-alpha.com/kolibri/

Independent evaluators

Evaluator harnesses are distinct from vendor measurements. Missing coverage is not a failed test.

No independent evaluator results are recorded for this model.

Read how we select and source scores or the comparison guide.