Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › granite-4.0 family

granite-4.0-h-small

Hugging Face ↗ Follow changes

ibm-granite Apache 2.0 out 13 months ago

32.2B parameters (10 of 72 experts per token), text in, text out, 128K tokens of context.

Verified 2026-10-25 Rank #688 in LLMs · score 72.4 method v1 Sources 3 Downloads 30d 26.2k · #437 of 619 tracked

Changed this week: 1 new quantized build on the Hub (37 in all).2026-10-24

Can I run it?19GB at INT4Estimatefits a 24 GB cardmethod
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24
Do people use it?26.2kdownloads, 30 days#437 of 619 we track · below the median+0 likes, downloads +5% this week311 likesHugging Face2026-10-24

Worth reading

Picked by rule from 4 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 13 months ago

IBM Granite 4.0

IBM launches Granite 4.0, a new family of open language models featuring a hybrid Mamba/transformer architecture designed for enterprise efficiency.

  • Granite 4.0-H-Small exceeds all open weight models on IFEval except Llama 4 Maverick, a 402B-parameter model over 12 times its size.
  • Granite 4.0 models are compatible with AMD Instinct MI-300X GPUs, enabling even further reduction of their memory footprint.
ibm.com 2025-10-02 10 HN pts · 3 comments 2.7k words about the granite 4.0 family family

4 best of 4 tracked

The 1 pick above, plus these 3. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. ArticleGranite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documentshuggingface.co · 2026-04-01
  2. ArticleIBM Granite 4.0 Tiny Previewibm.com · 2025-05-05 · 4 HN pts · official
  3. ArticleGranite 4.0 Nano: Just how small can you go?huggingface.co · 2025-10-28

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 37 quantized builds and 7 Spaces.

  • As a service0no API provider listed
  • On own hardware37quantized builds
  • Built on1338 fine-tunes · 125 adapters · 0 merges
  • In apps7Hugging Face Spaces

Top Spaces running it: akhaliq/granite-4.0-h-small ♥ 5 · TobDeBer/Granite4Family ♥ 4

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 64 GB Mac (unified memory) at INT4

About 19 GB at INT4, 39 GB at FP8, 77 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 19 GB of weights (estimate; KV cache and activations extra)INT419 GBFP8: about 39 GB of weights (estimate; KV cache and activations extra)FP839 GBBF16: about 77 GB of weights (estimate; KV cache and activations extra)BF1677 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 19 GB of weights (estimate; KV cache and activations extra)INT419 GBFP8: about 39 GB of weights (estimate; KV cache and activations extra)FP839 GBBF16: about 77 GB of weights (estimate; KV cache and activations extra)BF1677 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

2 dated articles and 10 repo commits so far

Teal is the weights (2025-09-16), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.

JunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-11-03: 2 repo commits2025-10-07: 1 repo commit2025-10-01: 1 repo commit2025-09-30: 2 repo commits2025-09-27: 1 repo commit2025-09-24: 1 repo commit2025-09-22: 1 repo commit2025-09-16: 1 repo commitIBM Granite 4.0 · 2025-10-02IBM Granite 4.0 Tiny Preview · 2025-05-05Weights 2025-09-16 · 2025-09-16Weights 2025-09-16now
JunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-11-03: 2 repo commits2025-10-07: 1 repo commit2025-10-01: 1 repo commit2025-09-30: 2 repo commits2025-09-27: 1 repo commit2025-09-24: 1 repo commit2025-09-22: 1 repo commit2025-09-16: 1 repo commitIBM Granite 4.0 · 2025-10-02IBM Granite 4.0 Tiny Preview · 2025-05-05Weights 2025-09-16 · 2025-09-16Weights 2025-09-16now

What people built on it

170 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

170 derivatives on the Hub, most of them adapters

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

37 quantized (Hugging Face model tree)8 fine-tunes (Hugging Face model tree)125 adapters (Hugging Face model tree)37 quantized8 fine-tunes125 adapters
37 quantized (Hugging Face model tree)8 fine-tunes (Hugging Face model tree)125 adapters (Hugging Face model tree)37 quantized8 fine-tunes125 adapters
GGUF unsloth/granite-4.0-h-small-GGUF for llama.cpp, Ollama, LM Studio 145.9k downloads MLX lmstudio-community/granite-4.0-h-small-MLX-8bit for Apple silicon 38.1k downloads FP8 ibm-granite/granite-4.0-h-small-FP8 for vLLM, SGLang 103.6k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

8 rows

Benchmarks 8 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (8)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
BenchmarkValueSetting / notesReported byDate
author-reported AGI EVAL70.630-shot, CoTmodel author2026-10-07
author-reported AlpacaEval · Chat and follow instructions42.48—model author2026-10-07
author-reported BBH · Reason through hard problems81.623-shot, CoTmodel author2026-10-07
author-reported GPQA Diamond · Reason through hard problems40.630-shot, CoTmodel author2026-10-07
author-reported IFEval · Chat and follow instructions89.87Instruct, Strictmodel author2026-10-07
author-reported IFEval · Chat and follow instructions85.22Prompt, Strictmodel author2026-10-07
author-reported MMLU · Reason through hard problems78.445-shotmodel author2026-10-07
author-reported MMLU-Pro · Reason through hard problems55.475-shot, CoTmodel author2026-10-07

What the card says

1 cited sentence

Summary generated, cited

Granite-4.0-H-Small is a 32.2B parameter mixture of experts instruct model designed for general instructions and business applications.[1][2][3]

  1. parameters human
  2. is mixture of experts
  3. “designed to respond to general instructions”

Generated 2026-10-07 15:14 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

6 use cases

What it is good at 6 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

22 fields

Listed in LLMs · MoE

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesGraniteMoeHybridForCausalLM
Model typegranitemoehybrid
Dtypebfloat16
Params by dtypeBF16: 32,207,337,984
Size bandL
MoEyes
Experts72
Experts per token10
Context length131,072 (config.json max_position_embeddings)
PapersarXiv:0000.00000
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · GraniteMoeHybridForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · GraniteMoeHybridForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for granite-4.0-h-small: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)77 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)39 GB · Single high-end GPU (<80 GB)
INT4 (0.5 bytes/param)19 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

40 variants

Lineage

Original model Family: ibm-granite/granite-4.0 170 derivatives on the Hub

quantized: 37finetune: 8adapter: 125

Top variants (40 shown, ranked by downloads)

Relation
Quant
Size
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
unsloth/granite-4.0-h-small-GGUF quantized GGUF — — apache-2.0 unsloth 51 145.9k
ibm-granite/granite-4.0-h-small-FP8 quantized -FP8 32.6B L apache-2.0 ibm-granite 5 103.6k
ibm-granite/granite-4.0-h-small-GGUF quantized GGUF — — apache-2.0 ibm-granite 24 43.1k
bartowski/ibm-granite_granite-4.0-h-small-GGUF quantized GGUF — — — bartowski 3 43.2k
lmstudio-community/granite-4.0-h-small-MLX-8bit quantized MLX 32.2B L apache-2.0 lmstudio-community 0 38.1k
cyankiwi/granite-4.0-h-small-AWQ-4bit quantized AWQ 9.69B L apache-2.0 cyankiwi 1 20.6k
lmstudio-community/granite-4.0-h-small-GGUF quantized GGUF — — apache-2.0 lmstudio-community 3 12k
mradermacher/granite-4.0-h-small-i1-GGUF quantized GGUF — — apache-2.0 mradermacher 0 10.6k
unsloth/granite-4.0-h-small finetune — 32.2B L apache-2.0 unsloth 4 7.6k
lmstudio-community/granite-4.0-h-small-MLX-6bit quantized MLX 32.2B L apache-2.0 lmstudio-community 1 4.2k
tiny-random/granite-moe-hybrid finetune — 5.86M S — tiny-random 0 3.4k
lmstudio-community/granite-4.0-h-small-MLX-4bit quantized MLX 32.2B L apache-2.0 lmstudio-community 0 3.3k
RedHatAI/granite-4.0-h-small-FP8-dynamic quantized -FP8 32.2B L apache-2.0 RedHatAI 2 2.6k
mradermacher/granite-4.0-h-small-GGUF quantized GGUF — — apache-2.0 mradermacher 1 2.6k
DevQuasar/ibm-granite.granite-4.0-h-small-GGUF quantized GGUF — — — DevQuasar 2 1.9k
lmstudio-community/granite-4.0-h-small-MLX-5bit quantized MLX 32.2B L apache-2.0 lmstudio-community 0 1.9k
unsloth/granite-4.0-h-small-FP8-Dynamic quantized -FP8 32.6B L apache-2.0 unsloth 3 1.5k
unsloth/granite-4.0-h-small-bnb-4bit quantized -bnb 32.2B L apache-2.0 unsloth 2 1.3k
RedHatAI/granite-4.0-h-small-FP8-block quantized -FP8 32.2B L apache-2.0 RedHatAI 0 1.3k
unsloth/granite-4.0-h-small-unsloth-bnb-4bit quantized -bnb 32.2B L apache-2.0 unsloth 3 1.1k
ExaltedSlayer/ibm-granite-4.0-h-small-mlx-mxfp4 quantized mlx 32.2B L apache-2.0 ExaltedSlayer 1 1.2k
ggml-org/granite-4.0-h-small-Q8_0-GGUF quantized Q8_0 — — apache-2.0 ggml-org 0 1k
noctrex/Granite-4.0-H-Small-MXFP4_MOE-GGUF quantized GGUF — — — noctrex 1 919
yujiepan/granite-4.0-h-tiny-random finetune — 5.86M S — yujiepan 0 963
taronaeo/Granite-4.0-H-Small-BE-GGUF quantized GGUF — — apache-2.0 taronaeo 0 946
amd/granite-4.0-h-small-fp8 quantized -fp8 32.2B L apache-2.0 amd 0 922
mlx-community/granite-4.0-h-small-4bit quantized -4bit 32.2B L apache-2.0 mlx-community 0 893
cyankiwi/granite-4.0-h-small-AWQ-8bit quantized AWQ 13.1B L apache-2.0 cyankiwi 1 769
yujiepan/granite-moe-hybrid-tiny-random finetune — 5.86M S — yujiepan 0 676
Myric/granite-4.0-h-small-APEX-GGUF quantized GGUF — — apache-2.0 Myric 0 624
mlx-community/granite-4.0-h-small-8bit quantized -8bit 32.2B L apache-2.0 mlx-community 0 583
MichielBuisman/granite-4.0-h-small-dutch-calibrated-gguf quantized gguf — — apache-2.0 MichielBuisman 0 399
AmpereComputing/granite-4.0-h-small-gguf quantized gguf — — — AmpereComputing 1 320
ethicalabs/granite-4.0-h-small-MLX quantized MLX 32.2B L apache-2.0 ethicalabs 0 304
amd/granite-4.0-h-small-w4a16-llmcompressor quantized -w4a16 32.2B L apache-2.0 amd 0 291
Govlens/granite-4.0-h-small-Q4_K_M-GGUF quantized GGUF — — apache-2.0 Govlens 0 270
platforms-aisg/granite-4.0-h-small-mlx-8Bit quantized mlx 32.2B L apache-2.0 platforms-aisg 0 268
allura-forge/claumba-average-sft finetune — 32.6B L apache-2.0 allura-forge 2 148
platforms-aisg/granite-4.0-h-small-mlx-4Bit quantized mlx 32.2B L apache-2.0 platforms-aisg 0 241
amd/granite-4.0-h-small-w8a8-llmcompressor quantized -w8a8 32.2B L apache-2.0 amd 0 234

Repository changelog

10 commits
changed in the last 7 daysOther: 2026-10-24all changes

Repository changelog 10 commits

DateCommitSHA
2025-11-03 model card update b8c0982
2025-11-03 Add MOT badge (#8) 88dedff
2025-10-07 adding default system prompt ba01e5b
2025-10-01 model card update 6981d0f
2025-09-30 model card update fc54b1f
2025-09-30 model card update 2aad1c3
2025-09-27 model card update f3a5bdc
2025-09-24 model card upload 9aa0a43
2025-09-22 model upload 76b0d8d
2025-09-16 initial commit a9d3527

Articles and coverage

4 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official2

IBM Granite 4.0
ibm.com2025-10-02officialauthorityabout the granite 4.0 familyHN 10 pts · 3 comments2,656 words
IBM launches Granite 4.0, a new family of open language models featuring a hybrid Mamba/transformer architecture designed for enterprise efficiency. Released under an Apache 2.0 license, these models are the first open models to receive ISO 42001 certification and offer over 70% reduction in RAM for long inputs. The hybrid architecture combines Mamba-2 layers with transformer attention, and IBM offers an uncapped indemnity for third party IP claims alongside a bug bounty program for security flaws.written, reviewed, cited
  • Granite 4.0-H-Small exceeds all open weight models on IFEval except Llama 4 Maverick, a 402B-parameter model over 12 times its size.
  • Granite 4.0 models are compatible with AMD Instinct MI-300X GPUs, enabling even further reduction of their memory footprint.
IBM Granite 4.0 Tiny Preview
ibm.com2025-05-05officialauthorityabout the granite 4.0 familyHN 4 pts · 0 comments1,604 words
IBM Granite 4.0 Tiny Preview is a preliminary open source model available under an Apache 2.0 license. It utilizes a hybrid Mamba-2/Transformer architecture with 7B total parameters and 1B active parameters at inference time. The model is pending support in Hugging Face transformers and vLLM before official local platform partner availability.written, reviewed, cited
  • Granite 4.0 Tiny Preview runs several concurrent 128K context sessions on consumer GPUs under $350 USD.
  • The model has only seen 2.5T of a planned 15T or more training tokens but rivals Granite 3.3 2B Instruct.
All coverage tracked for this model (4) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
Granite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documentshuggingface.co2026-04-012026-10-07hn
Granite 4.0 Nano: Just how small can you go?huggingface.co2025-10-282026-10-07hn
IBM Granite 4.0ibm.com official2025-10-022026-10-07hn
IBM Granite 4.0 Tiny Previewibm.com official2025-05-052026-10-07hn

Similar models

3 models

Similar models

  • GLM-4.7-Flash · 31.2B
    same size band (L), also MoE, same task (text-generation), more providers, permissive license
  • Qwen3-Coder-30B-A3B-Instruct · 30.5B
    same size band (L), also MoE, same task (text-generation), more providers, permissive license
  • Qwen3-30B-A3B · 30.5B
    same size band (L), also MoE, same task (text-generation), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:42 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:14 UTC · Last verified 2026-10-25

Follow granite-4.0-h-small

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.