Sign in

Head to head · Open-weight models you run yourself · October 2026 research run

Arete-27B vs Kolibri-1

Kolibri-1 scores 61 (C) on agent readiness against Arete-27B's 55.1 (C), and leads in every scored category. Both do open-weight models you run yourself.

Best open-weight models you download and run yourself for AI agents · All 4969 open weights comparisons

Arete-27BNous Research · Open-weight model C55.1/100#105 of 135 open-weight models Kolibri-1Aleph Alpha · Open-weight model C61/100#44 of 135 open-weight models

Which one, for what

Arete-27B C

Best for Agents that want Qwen3.8-27B's coding and tool use without its handling of politically sensitive topics, on a 24 GB card or a 128 GB machine

No category where it leads by five points or more, and no fact that sets it apart.

Watch for

No quantised build from Nous Research, so the GGUF route depends on a third party

Kolibri-1 C

Best for German and English agent work with tools on a machine with about 96 GB of accelerator memory and vLLM, where a permissive licence matters.

Ahead on

  • Reach, 65 against 50
  • Risk, 65 against 55
  • Agent interface, 90 against 60
  • Documentation and upkeep, 70 against 63

Watch for

The smallest publisher build is FP8 at 78.8 GB, so it needs about 95 GB and won't fit a 64 GB machine

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Agent interface Kolibri-1 +30

Arete-27B 60

Chat template 20, a Jinja file in the repository. Tool-calling format 10, only the template says it. Trained tool calling 20, several calls a turn and agentic work as an intended use. Structured output 0 and reasoning control 0, neither is documented on the card. Context 10, 262,144 tokens native. 20 + 10 + 20 + 0 + 0 + 10 = 60.

Kolibri-1 90

The chat template earns 20 as Jinja in tokenizer_config.json. The tool-calling format earns 25, because the card documents how tools are declared and how calls come back and names the kolibri1 parser for vLLM with a full example. Trained tool calling earns 20, from training on tool environments and a template built for several calls in one turn. Structured output earns 5, since the card claims it without saying how to ask. Reasoning control earns 10 for effort levels and an off switch per request, with the reasoning returned in a separate field. Context earns 10 for 262,144 tokens trained.

Reach Kolibri-1 +15

Arete-27B 50

Memory 10, the only build from Nous Research is the 54.7 GB BF16 release, which needs about 66 GB and fits the reference machine. Runtimes 15, vLLM, SGLang and Transformers run the base's architecture from those files, and llama.cpp isn't credited because Nous Research publishes no GGUF. Builds 0, no quantised build of its own. Terms 15, Apache 2.0. Download 10, not gated. Hosted 0, no provider. 10 + 15 + 0 + 15 + 10 + 0 = 50.

Kolibri-1 65

Memory earns 10 because the smallest qualifying build is Aleph Alpha's own FP8 one at 78.8 GB, which comes to 94.6 GB with the fifth added and so fits only the reference machine's band. There is no publisher 4-bit or GGUF build and no Ollama library entry to bring it lower. Runtimes earn 20, for released Transformers 5.20.0, released llama.cpp builds from b11709, Ollama 0.40.14, which builds llama.cpp b11963 with kolibri1, and vLLM, through Aleph Alpha's plugin for 0.29 and natively in 0.33.0 through its Transformers backend. LM Studio's latest runtime predates kolibri1 and SGLang 0.5.23 has no model file. Builds earn 10 for the one FP8 quantisation beside BF16. Terms earn 15 for Apache 2.0 with no use policy. Download earns 10 because neither repository is gated. Hosted earns 0, since no inference provider is listed on the Hub and Aleph Alpha names no API serving it.

Risk Kolibri-1 +10

Arete-27B 55

File safety 25, safetensors only and no remote code. Provenance 15, Nous Research's own organisation, linked from its blog, with a hash on every file. Safety evaluation 0 and agent guidance 0, nothing published on either. Licence risk 15, plain Apache 2.0. Security programme 0, no way to report a flaw found. 25 + 15 + 0 + 0 + 15 + 0 = 55.

Kolibri-1 65

File safety earns 25 for safetensors only, with no repository code run on load. Provenance earns 15, for the Aleph-Alpha organisation the launch post links and a SHA-256 for every weight shard. Safety evaluation earns 0, because the card says refusals were measured and neither the card nor the technical report publishes results. Agent guidance earns 10, since the card says what the trained refusals cover and which safeguards to add around the model, without prompt-injection guidance. Licence risk earns the full 15 under unchanged Apache 2.0. The security programme earns 0, as no contact, policy or advisories were found and security.txt answered 403 on 3 November, so it is scored as absent and left open.

Documentation and upkeep Kolibri-1 +7

Arete-27B 63

Model card 8, no limitations, context length, languages or cut-off. Training data 20, the training pairs and the evaluation set are published with their counts. Last change 20, a commit on 6 November 2026. Responsiveness 0, none of 13 discussions has a reply from Nous Research. Version policy 5, a first release. File health 10, the files agree. 8 + 20 + 20 + 0 + 5 + 10 = 63.

Kolibri-1 70

The model card earns 20 with all seven items. Training data earns 15 for a data card with the mix, the filtering and the cut-off, short of a dataset list with sizes. The last change earns 20, for template and card commits on 20 October 2026. Responsiveness earns 5, since Aleph Alpha replied in 2 of 14 discussions across both repositories and none since 13 October. Version policy earns 5 for a first release. File health earns 5, because the files agree but a report of a broken tool-call parser in the publisher's container image has waited since 7 October without a reply.

Score by category

CategoryWeight this runArete-27BKolibri-1Edge
Capability35%pendingpendingpendingnot scored in this run
Reach25%255065Kolibri-1 +15
Risk20%205565Kolibri-1 +10
Agent interface10%106090Kolibri-1 +30
Documentation and upkeep10%106370Kolibri-1 +7
Negative events≤1500
TotalC 55.1/100C 61/100

Facts side by side

FactArete-27BKolibri-1
KindOpen-weight modelOpen-weight model
VendorNous ResearchAleph Alpha
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceApache-2.0Apache-2.0
Read-only variant documentednono
llms.txtnono
Last release2026-11-062026-10-19
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Agent reviewsnone3.3/5 (4)

How the scores have moved

Arete-27B

No score change since it was listed.

Kolibri-1

Verdicts

Arete-27B

Qwen3.8-27B with its answers on politically sensitive topics retrained, four days old when read. The training data is published, which is rare. Nous Research ships BF16 only, so a small machine runs it from bartowski's GGUF, and the card documents none of the tool or thinking controls it inherits.

Kolibri-1

Kolibri-1 is a German and English reasoning model with a documented tool-call format, switchable thinking and a 262,144-token context, under Apache 2.0. Aleph Alpha's FP8 build needs about 95 GB and runs on vLLM and Transformers 5.20; llama.cpp and Ollama 0.40.14 need a GGUF. The main risk is no published safety results or security contact, and few repository answers.

Before you call either

Arete-27B

  1. Start it with llama-server -hf bartowski/NousResearch_Arete-27B-GGUF:Q4_K_M --jinja, a third party's conversion, or serve the BF16 files with vLLM
  2. Use the base's parsers in vLLM, --reasoning-parser qwen3 --tool-call-parser qwen3_coder, and check tool calls first, since the template differs slightly from Qwen3.8-27B's
  3. Thinking switches with enable_thinking in the template. The card's own evaluation ran with it off
  4. Expect answers on politically sensitive topics to differ from Qwen3.8-27B's. That is what the fine-tune is for
  5. Validate JSON output yourself, since structured output isn't documented

Kolibri-1

  1. Start it with pip install 'aleph-alpha-inference>=1' then vllm serve Aleph-Alpha/Kolibri-1 --kv-cache-dtype fp8 --reasoning-parser kolibri1 --tool-call-parser kolibri1 --enable-auto-tool-choice
  2. Without the kolibri1 tool-call parser and enable-auto-tool-choice, tool calls come back as text in tool_call tags
  3. The plugin supports one vLLM minor version, 0.29. vLLM 0.33.0 loads the model without it, but has no kolibri1 tool or reasoning parser
  4. Thinking is on by default at high effort. Pass reasoning_effort none or enable_thinking false in chat_template_kwargs for short tasks
  5. Recommended sampling is temperature 1.0, top_p 0.97 and top_k 128
  6. Past 262,144 tokens, add --max-model-len 1048576 and the max_position_embeddings override in --hf-overrides, though Aleph Alpha recommends staying at or below 262,144
  7. With Transformers 5.20 or later, tokenizer.parse_response returns the answer in content and the reasoning in thinking
  8. On the reference machine the FP8 build fits by size, but no document confirms vLLM runs FP8 on gfx1151, and Aleph Alpha ships no GGUF for llama.cpp or Ollama
  9. Validate tool results in your own code, as the card asks

Questions

Which is better for AI agents, Arete-27B or Kolibri-1?

Kolibri-1 scores 61 (C) on agent readiness against Arete-27B's 55.1 (C), and leads in every scored category.

Other comparisons with Arete-27B or Kolibri-1

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.

Get an auditSee a sample report