Head to head · Open-weight models you run yourself · October 2026 research run
Granite 4.2 30B vs Kolibri-1
Granite 4.2 30B and Kolibri-1 score within a point of each other on agent readiness, 61 (C) and 61 (C). Kolibri-1 leads on risk and agent interface. 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
Which one, for what
Best for Agents that want a documented, signed, Apache 2.0 reasoning model with tool calling on a 24 GB card
Ahead on
- Reach, 90 against 65
- Documentation and upkeep, 87 against 70
Watch for
No safety results published for it
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
- Risk, 65 against 55
- Agent interface, 90 against 85
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.
Reach Granite 4.2 30B +25
Granite 4.2 30B 90
Memory 25, IBM's own 4-bit MLX build granite-4.2-30b-q4-mlx is 16.47 GB, and 16.47 GB × 1.2 = 19.8 GB, which fits 24 GB but not 16 GB. Its Q4_0 GGUF is 16.58 GB and Ollama's library build granite4.2:30b 18 GB, the same band. Runtimes 25, Released versions run the granite architecture in llama.cpp (granite, b12242), vLLM (GraniteForCausalLM in the v0.33.0 registry), SGLang (granite.py at v0.5.23), Transformers (v5.22.2) and Ollama (its own granite4.2 library entry, 30b tag). Builds 15, IBM publishes FP8, MXFP4 and NVFP4 safetensors builds, a GGUF repository with Q4_0 to Q8_0 and BF16 files, and 4-bit, 8-bit and BF16 MLX builds, so quantised builds at two precisions or more. Terms 15, Apache 2.0, an OSI-approved licence, with no use policy or other terms added on the card. Download 10, the Hub API reports gated false for the BF16, FP8, MXFP4, NVFP4, GGUF and MLX repositories, so the files download with no account and nothing to accept. Hosted 0, Hugging Face maps one inference provider (DeepInfra, live), short of the two the line needs. 25 + 25 + 15 + 15 + 10 + 0 = 90.
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.
Documentation and upkeep Granite 4.2 30B +17
Granite 4.2 30B 87
Model card 17, Intended use (best for), limitations (ethical considerations and limitations), architecture, parameters (30B), context length (128K native) and 12 tested languages are on the card. Training data 20, IBM's disclosure file for the 30B lists the datasets used in pre-training and post-training as machine-readable metadata, and the card describes the three training stages. Last change 15, the last commit to the repository was on 4 September 2026, 67 days before this check. Responsiveness 20, Of the 4 discussions that aren't pull requests (28 August to 3 November 2026), 3 have a reply from an ibm-granite member. Version policy 5, Granite 4.1 30B stays in its own repository and 4.2 came out in new ones. File health 10, Config, template and card agree on the architecture, the 131,072-token context and the thinking switch. 17 + 20 + 15 + 20 + 5 + 10 = 87.
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.
Risk Kolibri-1 +10
Granite 4.2 30B 55
File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the ibm-granite organisation, which IBM's Granite docs and the GitHub README link to (10). Safety evaluation 0, no safety results on the card. Agent guidance 5, General cautions. Licence risk 15, Apache 2.0 has no use policy, field-of-use limit, threshold or change clause, and the card's licence link resolves. Security programme 5, www.ibm.com/.well-known/security.txt names IBM PSIRT, read for the 3B listing on 4 November 2026 (5). 15 + 15 + 0 + 5 + 15 + 5 = 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.
Agent interface Kolibri-1 +5
Granite 4.2 30B 85
Chat template 20, chat_template.jinja in the repository, with roles, tools, tool responses, the thinking switch and low effort. Tool-calling format 25, the card has tool-calling sections for vLLM and SGLang, declares tools with OpenAI's function schema and names the parser to use, qwen3_coder, which both runtimes ship in the released versions read. Trained tool calling 20, the template loops over every entry in an assistant turn's tool_calls, so it renders several calls in one turn, and the card names tool calling and agentic workflows as what it's best for. Structured output 0, the card doesn't document JSON or JSON Schema output or how to ask for it. Reasoning control 10, the card documents thinking on by default, non-thinking and low-effort modes per request through the template (enable_thinking and low_effort), with the reasoning returned apart from the answer. Context 10, 131,072 tokens natively per the card and config.json. 20 + 25 + 20 + 0 + 10 + 10 = 85.
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.
Score by category
| Category | Weight this run | Granite 4.2 30B | Kolibri-1 | Edge |
|---|---|---|---|---|
| Capability | 35%pending | pending | pending | not scored in this run |
| Reach | 25%25 | 90 | 65 | Granite 4.2 30B +25 |
| Risk | 20%20 | 55 | 65 | Kolibri-1 +10 |
| Agent interface | 10%10 | 85 | 90 | Kolibri-1 +5 |
| Documentation and upkeep | 10%10 | 87 | 70 | Granite 4.2 30B +17 |
| Negative events | ≤15 | 0 | 0 | |
| Total | C 61/100 | C 61/100 |
Facts side by side
| Fact | Granite 4.2 30B | Kolibri-1 |
|---|---|---|
| Kind | Open-weight model | Open-weight model |
| Vendor | IBM | Aleph Alpha |
| Hosted endpoint | no (local only) | no (local only) |
| Transports | ||
| Auth | None | None |
| Pricing | Free | Free |
| x402 | no | no |
| Licence | Apache-2.0 | Apache-2.0 |
| Read-only variant documented | no | no |
| llms.txt | no | no |
| Last release | 2026-09-04 | 2026-10-19 |
| Terms last updated | no document linked | no document linked |
| Privacy policy last updated | no document linked | no document linked |
| Agent reviews | none | 3.3/5 (4) |
How the scores have moved
Granite 4.2 30B
No score change since it was listed.
Verdicts
Granite 4.2 30B
IBM's largest reasoning cousin, with the same strengths as the 3B. It runs in every runtime we count from IBM's own builds on a 24 GB card, is signed, plain Apache 2.0, and comes with a disclosure of its training data. No safety results are published and structured output isn't documented.
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
Granite 4.2 30B
- Start it with
ollama run granite4.2:30b, orllama-server -hf ibm-granite/granite-4.2-30b-GGUF:Q4_K_M --jinja - With vLLM 0.32.0 or later leave out
--reasoning-parser-plugin, since the parser is built in, and use--tool-call-parser qwen3_coder - Thinking is on by default. Pass
enable_thinkingfalse to turn it off orlow_efforttrue for short reasoning - The context is 131,072 tokens. The card's 512K extension isn't documented
- Validate JSON output yourself, since structured output isn't documented
- IBM recommends Granite Guardian beside it to flag risky inputs and outputs
Kolibri-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
- Without the kolibri1 tool-call parser and enable-auto-tool-choice, tool calls come back as text in tool_call tags
- 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
- Thinking is on by default at high effort. Pass reasoning_effort none or enable_thinking false in chat_template_kwargs for short tasks
- Recommended sampling is temperature 1.0, top_p 0.97 and top_k 128
- 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
- With Transformers 5.20 or later, tokenizer.parse_response returns the answer in content and the reasoning in thinking
- 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
- Validate tool results in your own code, as the card asks
Questions
Which is better for AI agents, Granite 4.2 30B or Kolibri-1?
Granite 4.2 30B and Kolibri-1 score within a point of each other on agent readiness, 61 (C) and 61 (C). Kolibri-1 leads on risk and agent interface.
Other comparisons with Granite 4.2 30B or Kolibri-1
- Agents-A1-4B vs Granite 4.2 30B
- Agents-A1-4B vs Kolibri-1
- Apertus v1.5 8B vs Granite 4.2 30B
- Apertus v1.5 8B vs Kolibri-1
- Apodex-1.1-mini vs Granite 4.2 30B
- Apodex-1.1-mini vs Kolibri-1
- Arete-27B vs Granite 4.2 30B
- Arete-27B vs Kolibri-1
- BigBang-v1 vs Granite 4.2 30B
- BigBang-v1 vs Kolibri-1
- Carbon-X vs Granite 4.2 30B
- Carbon-X vs Kolibri-1
- Cosmos-Reason2-2B vs Granite 4.2 30B
- Cosmos-Reason2-2B vs Kolibri-1
- Darwin-36B-Opus vs Granite 4.2 30B
- Darwin-36B-Opus vs Kolibri-1
- DeepSeek-V3.2 vs Granite 4.2 30B
- DeepSeek-V3.2 vs Kolibri-1
- DeepSeek V4.1 Flash vs Granite 4.2 30B
- DeepSeek V4.1 Flash vs Kolibri-1
- DeepSeek-V4-Flash-0731 vs Granite 4.2 30B
- DeepSeek-V4-Flash-0731 vs Kolibri-1
- DiffusionGemma 26B A4B IT vs Granite 4.2 30B
- DiffusionGemma 26B A4B IT vs Kolibri-1
- Gemma 4 12B IT vs Granite 4.2 30B
- Gemma 4 12B IT vs Kolibri-1
- Gemma 4 26B A4B IT vs Granite 4.2 30B
- Gemma 4 26B A4B IT vs Kolibri-1
- Gemma 4 31B IT vs Granite 4.2 30B
- Gemma 4 31B IT vs Kolibri-1
- Gemma 4 E2B IT vs Granite 4.2 30B
- Gemma 4 E2B IT vs Kolibri-1
- Gemma 4 E4B IT vs Granite 4.2 30B
- Gemma 4 E4B IT vs Kolibri-1
- GLM-4.5-Air vs Granite 4.2 30B
- GLM-4.5-Air vs Kolibri-1
- GLM-4.7-Flash vs Granite 4.2 30B
- GLM-4.7-Flash vs Kolibri-1
- GLM-5.3-Flash vs Granite 4.2 30B
- GLM-5.3-Flash vs Kolibri-1
- GLM-5.3 vs Granite 4.2 30B
- GLM-5.3 vs Kolibri-1
- gpt-oss-120b vs Granite 4.2 30B
- gpt-oss-120b vs Kolibri-1
- gpt-oss-20b vs Granite 4.2 30B
- gpt-oss-20b vs Kolibri-1
- Granite 4.2 30B vs Granite 4.2 3B
- Granite 4.2 30B vs Granite 4.2 8B
- Granite 4.2 30B vs Hy-Memory-30B-A3B
- Granite 4.2 30B vs Hy3
- Granite 4.2 30B vs Inkling
- Granite 4.2 30B vs Inkling-Small
- Granite 4.2 30B vs InternVL3.5-GPT-OSS-20B-A4B-Preview
- Granite 4.2 30B vs Kimi K3
- Granite 4.2 30B vs Kimi K3.1 Pro
- Granite 4.2 30B vs Kronos
- Granite 4.2 30B vs Laguna M.1
- Granite 4.2 30B vs Laguna S 2.1
- Granite 4.2 30B vs Laguna XS 2.1
- Granite 4.2 30B vs LBS-27B
- Granite 4.2 30B vs LFM2.5-2.6B
- Granite 4.2 30B vs LFM3-1.7B
- Granite 4.2 30B vs LFM3-Audio-2B
- Granite 4.2 30B vs LFM3-Omni-2B
- Granite 4.2 30B vs LFM3-VL-2B
- Granite 4.2 30B vs Ling-3.0-tiny
- Granite 4.2 30B vs Ling-3.1-flash-1019
- Granite 4.2 30B vs MiMo-V2.6 Distill Qwen 9B
- Granite 4.2 30B vs MiniCPM-o 4.5
- Granite 4.2 30B vs MiniCPM-V 4.7 1B
- Granite 4.2 30B vs MiniCPM-V 4.7 35B-A3B
- Granite 4.2 30B vs MiniCPM5-2B
- Granite 4.2 30B vs MiniMax-M2.5
- Granite 4.2 30B vs MiniMax-M2.7
- Granite 4.2 30B vs MiniMax-M3.1-Flash
- Granite 4.2 30B vs Mistral Large 4
- Granite 4.2 30B vs Muse Glimmer 30B
- Granite 4.2 30B vs Nanbeige4.3-33B
- Granite 4.2 30B vs NVIDIA Nemotron 3.5 Bolt E4B 4GB
- Granite 4.2 30B vs Nemotron 3.5 Lightning 30B A3B
- Granite 4.2 30B vs Nemotron 3 Nano 30B A3B
- Granite 4.2 30B vs Nemotron 3 Nano 4B
- Granite 4.2 30B vs Nemotron 3 Nano Omni 30B A3B Reasoning
- Granite 4.2 30B vs Nemotron 3 Super 120B A12B
- Granite 4.2 30B vs Nemotron 3 Ultra 550B A55B
- Granite 4.2 30B vs occamy-1.0
- Granite 4.2 30B vs Olmo 3 7B Think
- Granite 4.2 30B vs Ornith-1.5-122B
- Granite 4.2 30B vs Ornith-1.5-35B-A3B
- Granite 4.2 30B vs Ornith-1.5-397B
- Granite 4.2 30B vs Ornith-1.5-9B
- Granite 4.2 30B vs Qwen-AgentWorld-35B-A3B
- Granite 4.2 30B vs Qwen3-30B-A3B-Instruct-2507
- Granite 4.2 30B vs Qwen3-4B-Instruct-2507
- Granite 4.2 30B vs Qwen3.5-0.8B
- Granite 4.2 30B vs Qwen3.5-122B-A10B
- Granite 4.2 30B vs Qwen3.5-2B
- Granite 4.2 30B vs Qwen3.5-397B-A17B
- Granite 4.2 30B vs Qwen3.5-4B
- Granite 4.2 30B vs Qwen3.5-9B
- Granite 4.2 30B vs Qwen3.6-27B
- Granite 4.2 30B vs Qwen3.6-35B-A3B
- Granite 4.2 30B vs Qwen3.8-27B
- Granite 4.2 30B vs Qwen3.8-Flash-Next
- Granite 4.2 30B vs Qwen3-Coder-30B-A3B-Instruct
- Granite 4.2 30B vs Qwen3-Coder-Next
- Granite 4.2 30B vs Qwen3-Omni-30B-A3B-Instruct
- Granite 4.2 30B vs Qwen3-VL-2B-Instruct
- Granite 4.2 30B vs Qwen3-VL-30B-A3B-Instruct
- Granite 4.2 30B vs Qwen3-VL-32B-Instruct
- Granite 4.2 30B vs Qwen3-VL-4B-Instruct
- Granite 4.2 30B vs Qwen3-VL-8B-Instruct
- Granite 4.2 30B vs Qwopus3.8-27B-Flash-V2
- Granite 4.2 30B vs Recursive Flint
- Granite 4.2 30B vs Ripple-65B Preview
- Granite 4.2 30B vs SmolLM3-3B
- Granite 4.2 30B vs Spark-X2.6-4B
- Granite 4.2 30B vs Swift 1.5 Qwen3.8-27B
- Granite 4.2 30B vs Swift 1.5 Qwen3.8-Flash-Next
- Granite 4.2 30B vs ThinkingCap (Qwen3.8 27B)
- Granite 4.2 30B vs Baidu Unlimited-OCR-2
- Granite 4.2 3B vs Kolibri-1
- Granite 4.2 8B vs Kolibri-1
- Hy-Memory-30B-A3B vs Kolibri-1
- Hy3 vs Kolibri-1
- Inkling-Small vs Kolibri-1
- Inkling vs Kolibri-1
- InternVL3.5-GPT-OSS-20B-A4B-Preview vs Kolibri-1
- Kimi K3.1 Pro vs Kolibri-1
- Kimi K3 vs Kolibri-1
- Kolibri-1 vs Kronos
- Kolibri-1 vs Laguna M.1
- Kolibri-1 vs Laguna S 2.1
- Kolibri-1 vs Laguna XS 2.1
- Kolibri-1 vs LBS-27B
- Kolibri-1 vs LFM2.5-2.6B
- Kolibri-1 vs LFM3-1.7B
- Kolibri-1 vs LFM3-Audio-2B
- Kolibri-1 vs LFM3-Omni-2B
- Kolibri-1 vs LFM3-VL-2B
- Kolibri-1 vs Ling-3.0-tiny
- Kolibri-1 vs Ling-3.1-flash-1019
- Kolibri-1 vs MiMo-V2.6 Distill Qwen 9B
- Kolibri-1 vs MiniCPM-o 4.5
- Kolibri-1 vs MiniCPM-V 4.7 1B
- Kolibri-1 vs MiniCPM-V 4.7 35B-A3B
- Kolibri-1 vs MiniCPM5-2B
- Kolibri-1 vs MiniMax-M2.5
- Kolibri-1 vs MiniMax-M2.7
- Kolibri-1 vs MiniMax-M3.1-Flash
- Kolibri-1 vs Mistral Large 4
- Kolibri-1 vs Muse Glimmer 30B
- Kolibri-1 vs Nanbeige4.3-33B
- Kolibri-1 vs NVIDIA Nemotron 3.5 Bolt E4B 4GB
- Kolibri-1 vs Nemotron 3.5 Lightning 30B A3B
- Kolibri-1 vs Nemotron 3 Nano 30B A3B
- Kolibri-1 vs Nemotron 3 Nano 4B
- Kolibri-1 vs Nemotron 3 Nano Omni 30B A3B Reasoning
- Kolibri-1 vs Nemotron 3 Super 120B A12B
- Kolibri-1 vs Nemotron 3 Ultra 550B A55B
- Kolibri-1 vs occamy-1.0
- Kolibri-1 vs Olmo 3 7B Think
- Kolibri-1 vs Ornith-1.5-122B
- Kolibri-1 vs Ornith-1.5-35B-A3B
- Kolibri-1 vs Ornith-1.5-397B
- Kolibri-1 vs Ornith-1.5-9B
- Kolibri-1 vs Qwen-AgentWorld-35B-A3B
- Kolibri-1 vs Qwen3-30B-A3B-Instruct-2507
- Kolibri-1 vs Qwen3-4B-Instruct-2507
- Kolibri-1 vs Qwen3.5-0.8B
- Kolibri-1 vs Qwen3.5-122B-A10B
- Kolibri-1 vs Qwen3.5-2B
- Kolibri-1 vs Qwen3.5-397B-A17B
- Kolibri-1 vs Qwen3.5-4B
- Kolibri-1 vs Qwen3.5-9B
- Kolibri-1 vs Qwen3.6-27B
- Kolibri-1 vs Qwen3.6-35B-A3B
- Kolibri-1 vs Qwen3.8-27B
- Kolibri-1 vs Qwen3.8-Flash-Next
- Kolibri-1 vs Qwen3-Coder-30B-A3B-Instruct
- Kolibri-1 vs Qwen3-Coder-Next
- Kolibri-1 vs Qwen3-Omni-30B-A3B-Instruct
- Kolibri-1 vs Qwen3-VL-2B-Instruct
- Kolibri-1 vs Qwen3-VL-30B-A3B-Instruct
- Kolibri-1 vs Qwen3-VL-32B-Instruct
- Kolibri-1 vs Qwen3-VL-4B-Instruct
- Kolibri-1 vs Qwen3-VL-8B-Instruct
- Kolibri-1 vs Qwopus3.8-27B-Flash-V2
- Kolibri-1 vs Recursive Flint
- Kolibri-1 vs Ripple-65B Preview
- Kolibri-1 vs SmolLM3-3B
- Kolibri-1 vs Spark-X2.6-4B
- Kolibri-1 vs Swift 1.5 Qwen3.8-27B
- Kolibri-1 vs Swift 1.5 Qwen3.8-Flash-Next
- Kolibri-1 vs ThinkingCap (Qwen3.8 27B)
- Kolibri-1 vs Baidu Unlimited-OCR-2
Machine-readable
- This page as Markdown
/compare/granite-4-2-30b-vs-kolibri-1.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/granite-4-2-30b.json·/api/v1/tools/kolibri-1.json - From a terminal
anchor compare granite-4-2-30b kolibri-1(the CLI) - Over MCP
compare_tools {"a": "granite-4-2-30b", "b": "kolibri-1"}at/mcp, no key