Models / IBM/ Granite 4.1 8B

Granite 4.1 8B

IBM · released Apr 6, 2026 · ibm-granite/granite-4.1-8b

Input: text. Output: text.InputOutput
Type
Open weightsApache License 2.0
Params
8.8B
Context
131K

about 98K words of context

Our take

Written Sep 30, 2026

Granite 4.1 8B is a downloadable text model from IBM that you can run yourself or reach through a host, and its licence allows commercial use, changes and redistribution. Its measured quality sits near the bottom of the Arena boards, so pick it for cheap bulk text work rather than for anything judged on the answer itself.

Who should pick it

Use it for high-volume text work where the bill matters more than answer quality — classification, extraction, routing, bulk rewriting — or when you want to run a model on a single machine. Its request capacity leaves room for a long report or a stack of documents beside the question, though reliable recall across all of it is unverified in our data. Skip it if the work is judged on the answer itself: coding, creative writing, maths or instruction following, where it places near the bottom of its fields.

The case for it

  • Both listed hosts sit at the same low rate, so bulk text processing stays inexpensive whichever you pick.
  • A request capacity of 131072 tokens leaves room for a long report or a stack of documents beside the question, though reliable recall across all of it is unverified in our data.
  • Apache License 2.0 allows commercial use, changes and redistribution.

The case against it

  • Human preference for its answers is near the bottom of the field: 145th of 168 on Arena Text (overall) as of 25 Sep 2026, a board that records which answer people preferred rather than whether it was correct.
  • On building web applications it places 94th of 95 on Arena Code (WebDev) as of 25 Sep 2026, a board scored from human votes on web-app building tasks.
  • Coding, writing and maths all place in the lower part of their boards: 143rd of 168 on Arena Coding, 147th of 168 on Arena Creative Writing and 129th of 163 on Arena Maths, all as of 25 Sep 2026.
00

How good is it?

An open text model from IBM for general chat, though it trails most models on everyday questions, writing and code.

Less good at
  • getting answers to everyday questionsArena Text (overall) · 145th of 168
  • drafts, rewrites and editingArena Creative Writing · 147th of 168
  • writing and completing codeArena Coding · 143rd of 168

EverydayGeneral questions and everyday reasoning

1 of 5

Arena Text (overall)145th of 168 · 1304

Arena Hard Prompts 141st of 168Arena Maths 129th of 163

CodingWriting and fixing code on its own

1.5 of 5

Arena Coding143rd of 168 · 1353

Arena Code (WebDev) 94th of 95

AgenticPlanning, calling tools, staying on task

not measured

Not yet scored on Arena Agent.

WritingDrafting and rewriting prose

1 of 5

Arena Creative Writing147th of 168 · 1260

Arena Creative Writing is the only board that has scored it for this.

Other boards it appears on
Arena Instruction Following 145th of 168

These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done.

Every published score for this model7 scoresEvery figure we hold, from 7 boards, with who ran it and a link to the source — including the boards no rating above is built on.
1353source ↗
1260source ↗
1320source ↗
1317source ↗
1304source ↗
1191source ↗
01

Can you run it yourself?

Comfortable fit

A card many people ownFits in memory

GeForce RTX 4090 · 24 GB

Weights at 5.6 / 24 GBest
Spare memory15.5 GB spare
Usable context66K of 131K
Decode speed151 tok/sest

Room to spare. 15.5 GB spare means a 10% error in the size would not change the answer.

One step upFits in memory

GeForce RTX 5090 · 32 GB

Weights at 5.6 / 32 GBest
Spare memory23.5 GB spare
Usable context131K of 131K
Decode speed269 tok/sest

Room to spare. 23.5 GB spare means a 10% error in the size would not change the answer.

On a MacFits in memory

Apple M1 (8-core GPU) · 16 GB

Weights at 5.6 / 16 GBest
Spare memory4.7 GB spare
Usable context16K of 131K
Decode speed9 tok/sest

Room to spare. 4.7 GB spare means a 10% error in the size would not change the answer.

Your hardware
Checking your profile…

Memory use by level

Against a 24 GB card.

What is quantisation? →
recommended
5.6 GBest
Fits in memory
6.5 GBest
Fits in memory
9.7 GBest
Fits in memory
This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
GeForce RTX 3060 8GB8 GB5.6 GBest2KFits in memoryest
GeForce RTX 4060 8GB8 GB5.6 GBest2KFits in memoryest
Radeon RX 66008 GB5.6 GBest2KFits in memoryest
Android phone · 16 GB · 2024 or newer8 GB5.6 GBest4KFits in memory
GeForce RTX 3080 10GB10 GB5.6 GBest8KFits in memory
Arc B57010 GB5.6 GBest8KFits in memory
GeForce RTX 507012 GB5.6 GBest16KFits in memory
GeForce RTX 4070 SUPER12 GB5.6 GBest16KFits in memory
Arc B58012 GB5.6 GBest16KFits in memory
GeForce RTX 3060 12GB12 GB5.6 GBest16KFits in memory
GeForce RTX 5070 Ti16 GB5.6 GBest33KFits in memory
GeForce RTX 508016 GB5.6 GBest33KFits in memory
GeForce RTX 4080 SUPER16 GB5.6 GBest33KFits in memory
GeForce RTX 4070 Ti SUPER16 GB5.6 GBest33KFits in memory
Radeon RX 907016 GB5.6 GBest33KFits in memory
Radeon RX 9070 XT16 GB5.6 GBest33KFits in memory
GeForce RTX 5060 Ti 16GB16 GB5.6 GBest33KFits in memory
GeForce RTX 4060 Ti 16GB16 GB5.6 GBest33KFits in memory
Apple M1 (8-core GPU)16 GB5.6 GBest16KFits in memory
Radeon RX 7900 XT20 GB5.6 GBest66KFits in memory
GeForce RTX 309024 GB5.6 GBest66KFits in memory
GeForce RTX 3090 Ti24 GB5.6 GBest66KFits in memory
GeForce RTX 409024 GB5.6 GBest66KFits in memory
Radeon RX 7900 XTX24 GB5.6 GBest66KFits in memory
Apple M2 (10-core GPU)24 GB5.6 GBest66KFits in memory
Apple M3 (10-core GPU)24 GB5.6 GBest66KFits in memory
GeForce RTX 509032 GB5.6 GBest131KFits in memory
Apple M1 Pro (16-core GPU)32 GB5.6 GBest66KFits in memory
Apple M2 Pro (19-core GPU)32 GB5.6 GBest66KFits in memory
Apple M5 (10-core GPU)32 GB5.6 GBest66KFits in memory
Apple M4 (10-core GPU)32 GB5.6 GBest66KFits in memory
Apple M3 Pro (18-core GPU)36 GB5.6 GBest66KFits in memory
L40S48 GB5.6 GBest131KFits in memory
RTX 6000 Ada48 GB5.6 GBest131KFits in memory
Apple M5 Max (32-core GPU)64 GB5.6 GBest131KFits in memory
Apple M1 Max (32-core GPU)64 GB5.6 GBest131KFits in memory
Apple M4 Max (32-core GPU)64 GB5.6 GBest131KFits in memory
Apple M5 Pro (20-core GPU)64 GB5.6 GBest131KFits in memory
Apple M4 Pro (20-core GPU)64 GB5.6 GBest131KFits in memory
A100 80GB SXM80 GB5.6 GBest131KFits in memory
H100 80GB SXM80 GB5.6 GBest131KFits in memory
RTX PRO 6000 Blackwell96 GB5.6 GBest131KFits in memory
Apple M2 Max (38-core GPU)96 GB5.6 GBest131KFits in memory
Apple M1 Ultra (64-core GPU)128 GB5.6 GBest131KFits in memory
Apple M5 Max (40-core GPU)128 GB5.6 GBest131KFits in memory
Apple M4 Max (40-core GPU)128 GB5.6 GBest131KFits in memory
Apple M3 Max (40-core GPU)128 GB5.6 GBest131KFits in memory
NVIDIA DGX Spark (GB10)128 GB5.6 GBest131KFits in memory
Ryzen AI Max+ 395 (Radeon 8060S)128 GB5.6 GBest131KFits in memory
H200 141GB SXM141 GB5.6 GBest131KFits in memory
B200 (SXM 192GB)192 GB5.6 GBest131KFits in memory
Instinct MI300X192 GB5.6 GBest131KFits in memory
Apple M2 Ultra (76-core GPU)192 GB5.6 GBest131KFits in memory
Apple M3 Ultra (80-core GPU)512 GB5.6 GBest131KFits in memory
GeForce GTX 1660 SUPER6 GB5.6 GBestnot calculatedSpills to system RAM
Apple M1 (8-core GPU, 8GB unified)8 GB5.6 GBestnot calculatedSpills to system RAM
Apple M2 (8-core GPU, 8GB unified)8 GB5.6 GBestnot calculatedSpills to system RAM
iPhone 17 Pro6.6 GB5.6 GBestnot calculatedToo large
Android phone · 12 GB · 2023 or newer6 GB5.6 GBestnot calculatedToo large
iPhone 15 Pro4.4 GB5.6 GBestnot calculatedToo large
iPhone 164.4 GB5.6 GBestnot calculatedToo large
iPhone 16 Pro4.4 GB5.6 GBestnot calculatedToo large
iPhone 174.4 GB5.6 GBestnot calculatedToo large
Android phone · 8 GB · 2020–20224 GB5.6 GBestnot calculatedToo large
Android phone · 8 GB · 2023 or newer4 GB5.6 GBestnot calculatedToo large
iPhone 143.3 GB5.6 GBestnot calculatedToo large
iPhone 153.3 GB5.6 GBestnot calculatedToo large
Android phone · 6 GB3 GB5.6 GBestnot calculatedToo large
iPhone 132.2 GB5.6 GBestnot calculatedToo large
iPhone SE (3rd gen)2.2 GB5.6 GBestnot calculatedToo large
Android phone · 4 GB2 GB5.6 GBestnot calculatedToo large

Check against your own machine → · Where to rent it hosted →

02

Or rent it from someone else

Prices checked 27 days ago — each listing carries its own date.

Cheapest published offer

Cheapest of 2 live listings.

per 1M tokens
$0.050 in / $0.10 out
Context served
131K
Throughput
Not measured
Current provider offers with price, context and prompt-privacy answers
ProviderIn / out per 1M tokensContextThroughputTrains on promptsLogs promptsZero retention
OpenRouterOpenRouter's own listing$0.050 / $0.10checked 27 days ago131Knot measuredUnknownUnknownUnknown
CoreWeavebf16Through OpenRouter$0.050 / $0.10checked 27 days ago131Knot measuredNoNoUnknown

Across the 2 listings we hold: 1 says it does not train on prompts, 0 say they do and 1 does not say. 0 appear in the zero-retention registry we check; the rest are unknown to us.

What each host's API supports

From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.

API features per host
ProviderTool callingJSON outputStrict schema
OpenRouterOpenRouter's own listing✓✓✓
CoreWeavebf16Through OpenRouter✓✓✓

Tool calling: 2 of 2 listings say yes. JSON output: 2 of 2 listings say yes. Strict schema: 2 of 2 listings say yes.

03

Models people weigh against Granite 4.1 8B

04

When we formed this view

Recent changes

Sep 25, 2026BenchmarkScored 1353 on Arena Coding
What movedleaderboard
Sep 25, 2026BenchmarkScored 1260 on Arena Creative Writing
What movedleaderboard
Sep 25, 2026BenchmarkScored 1320 on Arena Hard Prompts
What movedleaderboard
Sep 25, 2026BenchmarkScored 1287 on Arena Instruction Following
What movedleaderboard
Sep 25, 2026BenchmarkScored 1317 on Arena Maths
What movedleaderboard
Sep 25, 2026BenchmarkScored 1304 on Arena Text (overall)
What movedleaderboard
Sep 25, 2026BenchmarkScored 1191 on Arena Code (WebDev)
What movedleaderboard
Jul 26, 2026ListedListed on LLMap
What movedfirst indexed by our pipeline
Apr 6, 2026AnnouncedGranite 4.1 8B announced by IBM

Each date is the day we first saw the change, or the day the maker announced it.

What we do not know about this model yet

  • We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
  • Nothing we hold says whether an endpoint streams, so we do not show it either way.
  • 1 of 2 listings does not say whether it trains on prompts.
  • We hold no batch or off-peak rate for any of its listings.
  • We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
05

Licence and identifiers

What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.

Licence

Apache License 2.0

Open, few conditionsCommercial use allowed

Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.

Identifiers

Architecture
Dense
Takes in, gives back
Text in, text out
Catalogue slug
ibm-granite-granite-4-1-8b

Machine-readable model card (omc.json) →

Something wrong on this page? Tell us