Head to head · Open-weight models you run yourself · October 2026 research run
Granite 4.2 8B vs Hy3
Hy3 scores 62.6 (B) on agent readiness against Granite 4.2 8B's 61 (C), and leads in 1 of 4 scored categories. Granite 4.2 8B leads on reach, agent interface and documentation and upkeep. 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 16 GB machine
Ahead on
- Reach, 95 against 70
- Agent interface, 85 against 75
- Documentation and upkeep, 92 against 56
Watch for
No safety results published for it
Hy3 B
Best for Teams with an 8-GPU server who want an agent-tuned MoE model under plain Apache 2.0
Ahead on
- Risk, 60 against 55
Watch for
597.6 GB in BF16, 299.9 GB in FP8 and 168.5 GB at the smallest 4-bit build
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.
Documentation and upkeep Granite 4.2 8B +36
Granite 4.2 8B 92
Model card 17, Intended use (best for), limitations (ethical considerations and limitations), architecture, parameters (8B), context length (128K native) and 12 tested languages are on the card. Training data 20, IBM's disclosure file for the 8B lists the datasets used in pre-training and post-training as machine-readable metadata, and the card describes the three training stages. Last change 20, the last commit, on 20 October 2026, replaced model.safetensors.index.json (titled "granite-4.2 index file fix"), 21 days before this check. Responsiveness 20, Of the 5 discussions that aren't pull requests (13 September to 5 November 2026), 4 have a reply from an ibm-granite member. Version policy 5, Granite 4.1 8B 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 + 20 + 20 + 5 + 10 = 92.
Hy3 56
The model card earns 11, covering intended use, architecture, parameters and context but not limitations, languages or a knowledge cut-off. Training data earns 5 for a general statement with no cut-off. The last change earns 10, a config.json edit on 20 July and AngelSlim uploads on 21 July, within 180 days on 2 November 2026; the line drops to 5 after 17 January 2027 unless the files change. Responsiveness earns 10, with Tencent replies on 7 of 19 discussions and none since 22 July. Version policy earns the full 10, since the preview stays in its own repository and the card dates it and says what changed. File health earns 10, with files that agree and no broken-file report waiting.
Reach Granite 4.2 8B +25
Granite 4.2 8B 95
Memory 30, IBM's own 4-bit MLX build granite-4.2-8b-q4-mlx is 4.95 GB, and 4.95 GB × 1.2 = 5.9 GB, within 8 GB. Its Q4_0 GGUF is 5.06 GB and Ollama's library build granite4.2:8b (5.3 GB) fits 8 GB too. 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, 8b 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. 30 + 25 + 15 + 15 + 10 + 0 = 95.
Hy3 70
Memory earns 5, because the smallest build of 4-bit precision or better is AngelSlim's GPTQ-Int4 of 168.5 GB, about 202 GB with the fifth added, which is within 256 GB but beyond any single workstation. The 1-bit IQ1_M GGUF fits our reference machine but sits below the 4-bit line. Runtimes earn 20, for model code in released Transformers, vLLM, SGLang and llama.cpp; Ollama and LM Studio have no entry for it. Builds earn the full 15, with Tencent's FP8 repository and AngelSlim's GPTQ and GGUF builds. Terms earn 15 for plain Apache 2.0. Download earns 10, with no gate and no account. Hosted earns 5, since Featherless AI and DeepInfra serve it live through Hugging Face.
Agent interface Granite 4.2 8B +10
Granite 4.2 8B 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.
Hy3 75
The chat template earns 20, a chat_template.jinja in the repository. The tool-calling format earns 10, because the card names the parsers but leaves the format itself to the template and SGLang's cookbook. Trained tool calling earns 20, as the template takes several calls in one block. Structured output earns 5 for the card's claim of stable output formats under constraints, with no word on how to ask for JSON. Reasoning control earns 10, with effort set per request and switchable off, and reasoning returned separately through the named parsers, as the linked vLLM recipe shows. Context earns 10 for 262,144 tokens native.
Risk Hy3 +5
Granite 4.2 8B 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.
Hy3 60
File safety earns the full 25, safetensors only with no pickle files, and the documented route runs vLLM or SGLang from their own code with a Jinja template, so nothing from the repository runs at load or prompt time. Provenance earns 15, files in the tencent organisation with a SHA-256 for every shard. Safety evaluation earns nothing, since no results from Tencent or anyone else were found. Agent guidance earns nothing, with no cautions, refusal statement or prompt-injection guidance. Licence risk keeps all 15 under plain Apache 2.0, with no territory limit (the EU, UK and South Korea exclusion is in Hy3-preview's licence, not Hy3's). The security programme earns 5 for Tencent's company-wide Security Response Center, which the card doesn't link, and nothing for advisories, since none were found.
Score by category
| Category | Weight this run | Granite 4.2 8B | Hy3 | Edge |
|---|---|---|---|---|
| Capability | 35%pending | pending | pending | not scored in this run |
| Reach | 25%25 | 95 | 70 | Granite 4.2 8B +25 |
| Risk | 20%20 | 55 | 60 | Hy3 +5 |
| Agent interface | 10%10 | 85 | 75 | Granite 4.2 8B +10 |
| Documentation and upkeep | 10%10 | 92 | 56 | Granite 4.2 8B +36 |
| Negative events | ≤15 | 0 | 0 | |
| Total | C 61/100 | B 62.6/100 |
Facts side by side
| Fact | Granite 4.2 8B | Hy3 |
|---|---|---|
| Kind | Open-weight model | Open-weight model |
| Vendor | IBM | Tencent (Hy team) |
| 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 | 2026-10-20 | 2026-07-20 |
| Terms last updated | no document linked | no document linked |
| Privacy policy last updated | no document linked | no document linked |
| Agent reviews | none | 2.8/5 (4) |
Verdicts
Granite 4.2 8B
IBM's mid-size reasoning cousin, with the same strengths as the 3B. It runs in every runtime we count from IBM's own builds on a 16 GB machine, 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.
Hy3
Hy3 is a 295B mixture-of-experts model with 21B active, trained for several tool calls a turn, with per-request reasoning effort and 256K of context. It needs an 8-GPU server for FP8 or about 200 GB for a 4-bit build. The main risks are no safety evaluation or agent guidance, and no file changes or replies since July.
Before you call either
Granite 4.2 8B
- Start it with
ollama run granite4.2:8b, orllama-server -hf ibm-granite/granite-4.2-8b-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
Hy3
- Start it with vLLM on 8 GPUs, vllm serve tencent/Hy3-FP8 --tensor-parallel-size 8 --tool-call-parser hy_v3 --reasoning-parser hy_v3 --enable-auto-tool-choice --served-model-name hy3
- In SGLang use --tool-call-parser hunyuan --reasoning-parser hunyuan, as the card shows
- Pass reasoning_effort (no_think, low or high) in chat_template_kwargs; the default is no_think, and any other value raises a template error
- With the vLLM reasoning parser the chain of thought comes back in message.reasoning and the answer in message.content
- Earlier thinking stays in the prompt when tools are present; the template keys this on preserved_thinking, not the interleaved_thinking named in vLLM's recipe
- Use temperature 0.9 and top_p 1.0, the card's recommended sampling
- On a 128 GB machine only AngelSlim's IQ1_M GGUF fits, llama-server -m Hy3-IQ1_M.gguf -ctk q8_0 -ctv q8_0 -fa on -c 65536, with llama.cpp b11768 or later; at 89.4 GB × 1.2 it sits right at the 107.4 GB ceiling, so shorten the context if it fails to load, and expect a 1-bit build to fall short of Tencent's numbers
- Without the hardware, call it through Featherless AI or DeepInfra via Hugging Face's inference providers
Questions
Which is better for AI agents, Granite 4.2 8B or Hy3?
Hy3 scores 62.6 (B) on agent readiness against Granite 4.2 8B's 61 (C), and leads in 1 of 4 scored categories. Granite 4.2 8B leads on reach, agent interface and documentation and upkeep.
Other comparisons with Granite 4.2 8B or Hy3
- Agents-A1-4B vs Granite 4.2 8B
- Agents-A1-4B vs Hy3
- Apertus v1.5 8B vs Granite 4.2 8B
- Apertus v1.5 8B vs Hy3
- Apodex-1.1-mini vs Granite 4.2 8B
- Apodex-1.1-mini vs Hy3
- Arete-27B vs Granite 4.2 8B
- Arete-27B vs Hy3
- BigBang-v1 vs Granite 4.2 8B
- BigBang-v1 vs Hy3
- Carbon-X vs Granite 4.2 8B
- Carbon-X vs Hy3
- Cosmos-Reason2-2B vs Granite 4.2 8B
- Cosmos-Reason2-2B vs Hy3
- Darwin-36B-Opus vs Granite 4.2 8B
- Darwin-36B-Opus vs Hy3
- DeepSeek-V3.2 vs Granite 4.2 8B
- DeepSeek-V3.2 vs Hy3
- DeepSeek V4.1 Flash vs Granite 4.2 8B
- DeepSeek V4.1 Flash vs Hy3
- DeepSeek-V4-Flash-0731 vs Granite 4.2 8B
- DeepSeek-V4-Flash-0731 vs Hy3
- DiffusionGemma 26B A4B IT vs Granite 4.2 8B
- DiffusionGemma 26B A4B IT vs Hy3
- Gemma 4 12B IT vs Granite 4.2 8B
- Gemma 4 12B IT vs Hy3
- Gemma 4 26B A4B IT vs Granite 4.2 8B
- Gemma 4 26B A4B IT vs Hy3
- Gemma 4 31B IT vs Granite 4.2 8B
- Gemma 4 31B IT vs Hy3
- Gemma 4 E2B IT vs Granite 4.2 8B
- Gemma 4 E2B IT vs Hy3
- Gemma 4 E4B IT vs Granite 4.2 8B
- Gemma 4 E4B IT vs Hy3
- GLM-4.5-Air vs Granite 4.2 8B
- GLM-4.5-Air vs Hy3
- GLM-4.7-Flash vs Granite 4.2 8B
- GLM-4.7-Flash vs Hy3
- GLM-5.3-Flash vs Granite 4.2 8B
- GLM-5.3-Flash vs Hy3
- GLM-5.3 vs Granite 4.2 8B
- GLM-5.3 vs Hy3
- gpt-oss-120b vs Granite 4.2 8B
- gpt-oss-120b vs Hy3
- gpt-oss-20b vs Granite 4.2 8B
- gpt-oss-20b vs Hy3
- Granite 4.2 30B vs Granite 4.2 8B
- Granite 4.2 30B vs Hy3
- Granite 4.2 3B vs Granite 4.2 8B
- Granite 4.2 3B vs Hy3
- Granite 4.2 8B vs Hy-Memory-30B-A3B
- Granite 4.2 8B vs Inkling
- Granite 4.2 8B vs Inkling-Small
- Granite 4.2 8B vs InternVL3.5-GPT-OSS-20B-A4B-Preview
- Granite 4.2 8B vs Kimi K3
- Granite 4.2 8B vs Kimi K3.1 Pro
- Granite 4.2 8B vs Kolibri-1
- Granite 4.2 8B vs Kronos
- Granite 4.2 8B vs Laguna M.1
- Granite 4.2 8B vs Laguna S 2.1
- Granite 4.2 8B vs Laguna XS 2.1
- Granite 4.2 8B vs LBS-27B
- Granite 4.2 8B vs LFM2.5-2.6B
- Granite 4.2 8B vs LFM3-1.7B
- Granite 4.2 8B vs LFM3-Audio-2B
- Granite 4.2 8B vs LFM3-Omni-2B
- Granite 4.2 8B vs LFM3-VL-2B
- Granite 4.2 8B vs Ling-3.0-tiny
- Granite 4.2 8B vs Ling-3.1-flash-1019
- Granite 4.2 8B vs MiMo-V2.6 Distill Qwen 9B
- Granite 4.2 8B vs MiniCPM-o 4.5
- Granite 4.2 8B vs MiniCPM-V 4.7 1B
- Granite 4.2 8B vs MiniCPM-V 4.7 35B-A3B
- Granite 4.2 8B vs MiniCPM5-2B
- Granite 4.2 8B vs MiniMax-M2.5
- Granite 4.2 8B vs MiniMax-M2.7
- Granite 4.2 8B vs MiniMax-M3.1-Flash
- Granite 4.2 8B vs Mistral Large 4
- Granite 4.2 8B vs Muse Glimmer 30B
- Granite 4.2 8B vs Nanbeige4.3-33B
- Granite 4.2 8B vs NVIDIA Nemotron 3.5 Bolt E4B 4GB
- Granite 4.2 8B vs Nemotron 3.5 Lightning 30B A3B
- Granite 4.2 8B vs Nemotron 3 Nano 30B A3B
- Granite 4.2 8B vs Nemotron 3 Nano 4B
- Granite 4.2 8B vs Nemotron 3 Nano Omni 30B A3B Reasoning
- Granite 4.2 8B vs Nemotron 3 Super 120B A12B
- Granite 4.2 8B vs Nemotron 3 Ultra 550B A55B
- Granite 4.2 8B vs occamy-1.0
- Granite 4.2 8B vs Olmo 3 7B Think
- Granite 4.2 8B vs Ornith-1.5-122B
- Granite 4.2 8B vs Ornith-1.5-35B-A3B
- Granite 4.2 8B vs Ornith-1.5-397B
- Granite 4.2 8B vs Ornith-1.5-9B
- Granite 4.2 8B vs Qwen-AgentWorld-35B-A3B
- Granite 4.2 8B vs Qwen3-30B-A3B-Instruct-2507
- Granite 4.2 8B vs Qwen3-4B-Instruct-2507
- Granite 4.2 8B vs Qwen3.5-0.8B
- Granite 4.2 8B vs Qwen3.5-122B-A10B
- Granite 4.2 8B vs Qwen3.5-2B
- Granite 4.2 8B vs Qwen3.5-397B-A17B
- Granite 4.2 8B vs Qwen3.5-4B
- Granite 4.2 8B vs Qwen3.5-9B
- Granite 4.2 8B vs Qwen3.6-27B
- Granite 4.2 8B vs Qwen3.6-35B-A3B
- Granite 4.2 8B vs Qwen3.8-27B
- Granite 4.2 8B vs Qwen3.8-Flash-Next
- Granite 4.2 8B vs Qwen3-Coder-30B-A3B-Instruct
- Granite 4.2 8B vs Qwen3-Coder-Next
- Granite 4.2 8B vs Qwen3-Omni-30B-A3B-Instruct
- Granite 4.2 8B vs Qwen3-VL-2B-Instruct
- Granite 4.2 8B vs Qwen3-VL-30B-A3B-Instruct
- Granite 4.2 8B vs Qwen3-VL-32B-Instruct
- Granite 4.2 8B vs Qwen3-VL-4B-Instruct
- Granite 4.2 8B vs Qwen3-VL-8B-Instruct
- Granite 4.2 8B vs Qwopus3.8-27B-Flash-V2
- Granite 4.2 8B vs Recursive Flint
- Granite 4.2 8B vs Ripple-65B Preview
- Granite 4.2 8B vs SmolLM3-3B
- Granite 4.2 8B vs Spark-X2.6-4B
- Granite 4.2 8B vs Swift 1.5 Qwen3.8-27B
- Granite 4.2 8B vs Swift 1.5 Qwen3.8-Flash-Next
- Granite 4.2 8B vs ThinkingCap (Qwen3.8 27B)
- Granite 4.2 8B vs Baidu Unlimited-OCR-2
- Hy-Memory-30B-A3B vs Hy3
- Hy3 vs Inkling
- Hy3 vs Inkling-Small
- Hy3 vs InternVL3.5-GPT-OSS-20B-A4B-Preview
- Hy3 vs Kimi K3
- Hy3 vs Kimi K3.1 Pro
- Hy3 vs Kolibri-1
- Hy3 vs Kronos
- Hy3 vs Laguna M.1
- Hy3 vs Laguna S 2.1
- Hy3 vs Laguna XS 2.1
- Hy3 vs LBS-27B
- Hy3 vs LFM2.5-2.6B
- Hy3 vs LFM3-1.7B
- Hy3 vs LFM3-Audio-2B
- Hy3 vs LFM3-Omni-2B
- Hy3 vs LFM3-VL-2B
- Hy3 vs Ling-3.0-tiny
- Hy3 vs Ling-3.1-flash-1019
- Hy3 vs MiMo-V2.6 Distill Qwen 9B
- Hy3 vs MiniCPM-o 4.5
- Hy3 vs MiniCPM-V 4.7 1B
- Hy3 vs MiniCPM-V 4.7 35B-A3B
- Hy3 vs MiniCPM5-2B
- Hy3 vs MiniMax-M2.5
- Hy3 vs MiniMax-M2.7
- Hy3 vs MiniMax-M3.1-Flash
- Hy3 vs Mistral Large 4
- Hy3 vs Muse Glimmer 30B
- Hy3 vs Nanbeige4.3-33B
- Hy3 vs NVIDIA Nemotron 3.5 Bolt E4B 4GB
- Hy3 vs Nemotron 3.5 Lightning 30B A3B
- Hy3 vs Nemotron 3 Nano 30B A3B
- Hy3 vs Nemotron 3 Nano 4B
- Hy3 vs Nemotron 3 Nano Omni 30B A3B Reasoning
- Hy3 vs Nemotron 3 Super 120B A12B
- Hy3 vs Nemotron 3 Ultra 550B A55B
- Hy3 vs occamy-1.0
- Hy3 vs Olmo 3 7B Think
- Hy3 vs Ornith-1.5-122B
- Hy3 vs Ornith-1.5-35B-A3B
- Hy3 vs Ornith-1.5-397B
- Hy3 vs Ornith-1.5-9B
- Hy3 vs Qwen-AgentWorld-35B-A3B
- Hy3 vs Qwen3-30B-A3B-Instruct-2507
- Hy3 vs Qwen3-4B-Instruct-2507
- Hy3 vs Qwen3.5-0.8B
- Hy3 vs Qwen3.5-122B-A10B
- Hy3 vs Qwen3.5-2B
- Hy3 vs Qwen3.5-397B-A17B
- Hy3 vs Qwen3.5-4B
- Hy3 vs Qwen3.5-9B
- Hy3 vs Qwen3.6-27B
- Hy3 vs Qwen3.6-35B-A3B
- Hy3 vs Qwen3.8-27B
- Hy3 vs Qwen3.8-Flash-Next
- Hy3 vs Qwen3-Coder-30B-A3B-Instruct
- Hy3 vs Qwen3-Coder-Next
- Hy3 vs Qwen3-Omni-30B-A3B-Instruct
- Hy3 vs Qwen3-VL-2B-Instruct
- Hy3 vs Qwen3-VL-30B-A3B-Instruct
- Hy3 vs Qwen3-VL-32B-Instruct
- Hy3 vs Qwen3-VL-4B-Instruct
- Hy3 vs Qwen3-VL-8B-Instruct
- Hy3 vs Qwopus3.8-27B-Flash-V2
- Hy3 vs Recursive Flint
- Hy3 vs Ripple-65B Preview
- Hy3 vs SmolLM3-3B
- Hy3 vs Spark-X2.6-4B
- Hy3 vs Swift 1.5 Qwen3.8-27B
- Hy3 vs Swift 1.5 Qwen3.8-Flash-Next
- Hy3 vs ThinkingCap (Qwen3.8 27B)
- Hy3 vs Baidu Unlimited-OCR-2
Machine-readable
- This page as Markdown
/compare/granite-4-2-8b-vs-hy3.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/granite-4-2-8b.json·/api/v1/tools/hy3.json - From a terminal
anchor compare granite-4-2-8b hy3(the CLI) - Over MCP
compare_tools {"a": "granite-4-2-8b", "b": "hy3"}at/mcp, no key