granite-4.2-8b

8.79B parameters, text in, text out, 128K tokens of context. 2 providers serve it from $0.060 per million input tokens.

Verified 2026-10-25 Rank #599 in LLMs · score 74.1 method v1 Sources 4 Downloads 30d 80.2k · #339 of 619 tracked

Changed this week: 5 new quantized builds on the Hub (59 in all).2026-10-24

Can I run it?5.3GB at INT4Estimatefits a 16 GB cardmethod

What does it cost?$0.060/1M incheapest of 2DeepInfrabf162026-10-25

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Do people use it?80.2kdownloads, 30 days#339 of 619 we track · below the median+3 likes, downloads −15% this week107 likesHugging Face2026-10-24

Worth reading

Picked by rule from 2 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 2 months ago

Granite 4.2 brings native reasoning to enterprise agents

IBM releases Granite 4.2, an open model family under an Apache 2.0 license designed for enterprise agents with native reasoning and tool use.

Granite 4.2 uses a speculative decoding layer to output text faster while serving more users.
Hirundo's machine unlearning technology targets undesirable outputs without the need to fully retrain the model.

research.ibm.com 2026-08-25 2 HN pts 1.1k words about the granite 4.2 family family

2 best of 2 tracked

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

ArticleIBM's new Granite 4.2 models ride the wave of interest in local LLMsarstechnica.com · 2026-08-26 · 4 HN pts

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

Prices for the same weights vary 2×

2 providers with a live price, from $0.060 to $0.10 per million input tokens; the median is $0.080. One dot per provider, cheapest in teal.

#	Cheapest 2	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.060	$0.25	128K	bf16	Oct 25
2	CoreWeave	$0.10	$0.15	128K	bf16	Oct 24

All 2 providers →

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.

Served by 2 API providers; heavily run locally (59 quantized builds).

As a service2CoreWeave, DeepInfra
On own hardware59quantized builds
Built on2618 fine-tunes · 8 adapters · 0 merges
In apps6Hugging Face Spaces

Top Spaces running it: abonaie222263/hf-openai-proxy ♥ 1 · Infergauge-load/llm-performance-and-load ♥ 1

Will it fit?

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

Fits a 16 GB card at INT4

About 5.3 GB at INT4, 11 GB at FP8, 21 GB at BF16: weights only, so leave room for the KV cache.

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 18 repo commits so far

Teal is the weights (2026-08-07), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

85 derivatives on the Hub, most of them quantized

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

GGUF ibm-granite/granite-4.2-8b-GGUF for llama.cpp, Ollama, LM Studio 636.2k downloads MLX lmstudio-community/granite-4.2-8b-MLX-8bit for Apple silicon 10k downloads FP8 ibm-granite/granite-4.2-8b-fp8 for vLLM, SGLang 5.6k downloads

All variants →

Reference

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

What the card says

3 cited sentences

Summary generated, cited

Granite-4.2-8B is a decoder-only dense transformer model with 8.79B parameters.^[1]^[2] The model is designed for reasoning, code generation, and tool calling using a built-in chain-of-thought mechanism.^[3] It is released under the Apache 2.0 license and supports a context window up to 512K.^[4]^[5]

parameters human
“Decoder-only Dense Transformer”
“Built-in <think>...</think> chain-of-thought”
license
“512K Context Window”

Generated 2026-10-07 15:12 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

7 use cases

What it is good at 7 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

21 fields

Listed in LLMs

Facts

granite-4.2-8b has 8.79B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2026-08-07, last modified 2026-10-20. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	GraniteForCausalLM
Model type	granite
Dtype	bfloat16
Params by dtype	BF16: 8,791,592,960
Size band	M
MoE	no
Context length	131,072 (config.json max_position_embeddings)
Languages	en, de, es, fr, ja, pt, ar, cs, it, ko, nl, zh
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · GraniteForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · GraniteForCausalLM · 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-4.2-8b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	21 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	11 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)	5.3 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Fine-tuned from ibm-granite/granite-4.1-8b-base (same organization) Family: ibm-granite/granite-4.2 85 derivatives on the Hub

Base model: ibm-granite/granite-4.1-8b-base

quantized: 59finetune: 18adapter: 8

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
ibm-granite/granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	ibm-granite	47	636.2k
bartowski/granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	bartowski	9	13.5k
lmstudio-community/granite-4.2-8b-MLX-8bit	quantized	MLX	8.79B	M	apache-2.0	lmstudio-community	1	10k
lmstudio-community/granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	lmstudio-community	2	5.7k
ibm-granite/granite-4.2-8b-fp8	quantized	-fp8	8.79B	M	apache-2.0	ibm-granite	2	5.6k
lmstudio-community/granite-4.2-8b-MLX-4bit	quantized	MLX	8.79B	M	apache-2.0	lmstudio-community	1	5k
mradermacher/granite-4.2-8b-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	2	4.7k
ibm-granite/granite-4.2-8b-nvfp4	quantized	nvfp4	5.31B	M	apache-2.0	ibm-granite	7	4k
ibm-granite/granite-4.2-8b-mxfp4	quantized	mxfp4	5.06B	M	apache-2.0	ibm-granite	2	2.5k
NANI-Nithin/granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	NANI-Nithin	1	1.8k
saidutta69/granite-4.2-8b-heretic	quantized	GGUF	8.79B	M	apache-2.0	saidutta69	0	1.8k
lmstudio-community/granite-4.2-8b-MLX-6bit	quantized	MLX	8.79B	M	apache-2.0	lmstudio-community	0	1.7k
ryugyosoft/granite-4.2-8b-onw	finetune	—	—	—	apache-2.0	ryugyosoft	0	1.6k
richardyoung/granite-4.2-8b-heretic	finetune	—	8.79B	M	apache-2.0	richardyoung	0	1.5k
ibm-granite/granite-4.2-8b-q4-mlx	quantized	mlx	8.79B	M	apache-2.0	ibm-granite	1	1.3k
mradermacher/granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	1.3k
nile-academy/granite-4.2-8b-gguf	quantized	gguf	—	—	apache-2.0	nile-academy	0	1.2k
rolandwhere/granite-4.2-8b-symbolic-cot	finetune	—	8.79B	M	apache-2.0	rolandwhere	5	925
webmp3/Sakura-granite-4.2-8b-GGUF	quantized	GGUF	—	—	apache-2.0	webmp3	1	1.1k
ibm-granite/granite-4.2-8b-q8-mlx	quantized	mlx	—	—	apache-2.0	ibm-granite	1	1k
xCloudinfo/Granite-4.2-8B-Uncensored-GGUF	quantized	GGUF	—	—	apache-2.0	xCloudinfo	0	1.1k
ibm-granite/granite-4.2-8b-bf16-mlx	finetune	-bf16	—	—	apache-2.0	ibm-granite	1	830
taronaeo/Granite-4.2-8B-BE-GGUF	quantized	GGUF	—	—	apache-2.0	taronaeo	0	832
darioooooo0o/granite-4.2-8b-G-Q5-GGUF	quantized	GGUF	—	—	apache-2.0	darioooooo0o	1	735
liodon-ai/granite-4.2-8b-imatrix-GGUF	quantized	imatrix	—	—	other	liodon-ai	0	603
salohcin714/granite-4.2-8b-bf16-mlx	finetune	-bf16	8.79B	M	apache-2.0	salohcin714	0	557
webAI-Official/granite-4.2-8B	quantized	GGUF	—	—	apache-2.0	webAI-Official	1	492
liodon-ai/granite-4.2-8b-ONNX	quantized	—	—	—	other	liodon-ai	0	466
ahmetggg/Drzeon-granite-4-2-8b-stem-finetune	adapter	GGUF	—	—	apache-2.0	ahmetggg	1	377
salohcin714/granite-4.2-8b-2bit-mlx	quantized	mlx	8.79B	M	apache-2.0	salohcin714	0	424
Jeethu/granite-4.2-8b-PARO	quantized	—	1.91B	S	apache-2.0	Jeethu	0	419
hirundo-io/granite-4.2-8b-hardened	finetune	—	8.79B	M	apache-2.0	hirundo-io	0	412
salohcin714/granite-4.2-8b-3bit-mlx	quantized	mlx	8.79B	M	apache-2.0	salohcin714	0	409
salohcin714/granite-4.2-8b-mxfp4-mlx	quantized	mxfp4	8.79B	M	apache-2.0	salohcin714	0	407
salohcin714/granite-4.2-8b-nvfp4-mlx	quantized	nvfp4	8.79B	M	apache-2.0	salohcin714	0	380
elBundinio/granite-4.2-8b-openvino	finetune	—	—	—	apache-2.0	elBundinio	0	361
salohcin714/granite-4.2-8b-5bit-mlx	quantized	mlx	8.79B	M	apache-2.0	salohcin714	0	359
livrrr/granite-4.2-8b-openvino	finetune	—	—	—	apache-2.0	livrrr	0	321
salohcin714/granite-4.2-8b-2bit-gptq-mlx	quantized	gptq	8.79B	M	apache-2.0	salohcin714	0	310
mohameddalii/coda-llm	finetune	—	8.79B	M	apache-2.0	mohameddalii	0	307

Repository changelog

18 commits

changed in the last 7 daysRepo commits: 2026-10-20, 2026-10-20, 2026-10-20Other: 2026-10-24all changes

Repository changelog 18 commits

Date	Commit	SHA
2026-10-20	granite-4.2 index file fix	a704214
2026-10-20	Delete __pycache__	c68f730
2026-10-20	granite-4.2 index file fix	6af7db3
2026-09-04	granite-4.2	f8de16c
2026-09-02	granite-4.2	7fce579
2026-08-25	Update Serving with SGLang section (#1)	41a8a2d
2026-08-25	Sync model card and signature from release staging	d554f3c
2026-08-24	Update README.md	313d025
2026-08-24	Update README.md	67efb21
2026-08-24	Update README.md	609b649
2026-08-19	granite-4.2	473b151
2026-08-18	granite-4.2	1cb039c
2026-08-13	granite-4.2	08e90a6
2026-08-11	granite-4.2	3778487
2026-08-07	granite-4.2	27edf44
2026-08-07	Update chat_template.jinja	28ad45f
2026-08-07	granite-4.2	eda3381
2026-08-07	initial commit	886b57a

All providers

2 providers · 2 endpoints

Where to run 2 providers · 2 endpoints

Hosted endpoints observed for granite-4.2-8b, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.06/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra hf, deepinfra, openrouter	ibm-granite/granite-4.2-8b	128K	bf16	$0.06	$0.25	34.1	324 ms	yes	live	2026-10-25
CoreWeave openrouter	ibm-granite/granite-4.2-8b	128K	bf16	$0.10	$0.15	—	—	yes	live	2026-10-24

Articles and coverage

2 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.

Official1

Granite 4.2 brings native reasoning to enterprise agents

research.ibm.com2026-08-25officialauthorityabout the granite 4.2 familyHN 2 pts · 0 comments1,085 words

IBM releases Granite 4.2, an open model family under an Apache 2.0 license designed for enterprise agents with native reasoning and tool use. Available in 3B, 8B, and 30B parameter sizes, the models were trained on 1 trillion tokens of synthetic code to enable step-by-step planning before acting. New Granite Speech 5.0 Turbo CTC models achieve a processing throughput of 12,600 RTFx on a single H200 GPU, making them suitable for high-throughput edge transcription.written, reviewed, cited

Granite 4.2 uses a speculative decoding layer to output text faster while serving more users.
Hirundo's machine unlearning technology targets undesirable outputs without the need to fully retrain the model.
Granite Speech 5.0 Turbo CTC models have no LLM backbone, keeping the model size minuscule while remaining efficient.

Analysis1

IBM's new Granite 4.2 models ride the wave of interest in local LLMs

arstechnica.com2026-08-26about the granite 4.2 familyHN 4 pts · 0 comments584 words

IBM has released the Granite 4.2 family of open-weight large language models designed for local deployment. The 8B and 30B variants undergo an agentic reinforcement-learning block to enable capabilities like using the terminal, searching the web, or using external tools, while the 3B model supports tools without specialized training. This reasoning-focused release offers a native 128,000-token context window but often incurs slower response times and higher compute demands compared to competitors like Nvidia's Nemotron.written, reviewed, cited

Similar models

3 models

Similar models

Llama-3.1-8B-Instruct · 8.03B
same size band (M), same task (text-generation), more providers
Qwen3-8B · 8.19B
same size band (M), same task (text-generation), more providers, permissive license
Meta-Llama-3-8B-Instruct · 8.03B
same size band (M), same task (text-generation)

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:40 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:12 UTC · Last verified 2026-10-25

Follow granite-4.2-8b

Watch this model

/feeds/models/ibm-granite--granite-4.2-8b.xml(RSS 2.0) · every change to granite-4.2-8b
/feeds/families/ibm-granite--granite-4.2.xml(RSS 2.0) · every change across the granite-4.2 family

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