Granite 4.0 350M
Oct 2025
Granite 4.0 350MCompare inon
Benchmarks of Granite 4.0 350M15
BenchmarkByScoreRank
Overall1
Knowledge4
Science2
Coding1
Legacy · saturated or no longer reported7
No benchmark matches
—per task?
self-host onlyprice
—output speed?
33Kcontext
Rankingin non-thinking mode?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-6 Lunalow×4.5—$0.005Function calling only with none · Chat Completions+3IOpenK2 Horizon 7Bthinking×4.3—$0.30 / $1.20Memory (4-bit): 0.2 GB → 4.5 GBBetterGranite 4.2 3BthinkingOpen+87%—$0.006Memory (4-bit): 0.2 GB → 1.8 GB
BetterGemma 4 E4BthinkingOpen+84%—$0.008Memory (4-bit): 0.2 GB → 4 GBBetterGPT-5.6 Lunalow×4.3—$0.01ModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Lunalow×4.5—$0.005Function calling only with none · Chat Completions+3Granite 4.2 3BthinkingOpen+87%—$0.006Memory (4-bit): 0.2 GB → 1.8 GBGemma 4 E4BthinkingOpen+84%—$0.008Memory (4-bit): 0.2 GB → 4 GBGPT-5.6 Lunalow×4.3—$0.01GPT OSS 20BhighOpen+86%—$0.011Memory (4-bit): 0.2 GB → 13 GBModelScoreCostPer taskWhat changes
IK2 Horizon 7BthinkingOpen×4.3—$0.30 / $1.20Memory (4-bit): 0.2 GB → 4.5 GBIK2 Horizon 3.7BthinkingOpen×3.2—$0.30 / $1.20Memory (4-bit): 0.2 GB → 2.5 GBGemma 4 12BthinkingOpen×2.9—$0.10 / $0.30Memory (4-bit): 0.2 GB → 6 GB
Qwen 3.5 9Bnon-thinkingOpen×2.7—$0.08 / $0.13Memory (4-bit): 0.2 GB → 4.8 GBQwen3.5 4BthinkingOpen×2.7—$0.03 / $0.15Memory (4-bit): 0.2 GB → 2.3 GBModelScoreCostPer taskWhat changes
Granite 4.2 30BthinkingOpen×2.6—$0.072Memory (4-bit): 0.2 GB → 18 GBGranite 4.2 8BthinkingOpen×2.3—$0.024Memory (4-bit): 0.2 GB → 4.4 GBGranite 4.2 3BthinkingOpen+87%—$0.006Memory (4-bit): 0.2 GB → 1.8 GBGranite 4.1 30BOpen+54%—$0.10 / $0.25Memory (4-bit): 0.2 GB → 17 GBGranite 4.1 8Bnon-thinkingOpen+36%——Memory (4-bit): 0.2 GB → 4.4 GBGranite 4.0 H Smallnon-thinkingOpen+25%—$0.05 / $0.22Memory (4-bit): 0.2 GB → 19 GBGranite 4.1 3BOpen+21%——Memory (4-bit): 0.2 GB → 1.7 GBGranite 4.0 H 1BOpensame——Memory (4-bit): 0.2 GB → 0.7 GBAll IBM models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · full quality✓ 16 GB · full quality✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam5.5#259
AA Omniscience Index-69.4#293
CritPt (physics)0#180
AA Intelligence Index 4.3.2≈4.8#327
AA Omniscience · Accuracy3.9#318
GPQA Diamond26.1#308
Coding
LiveCodeBench · AA archived2.4#150
Agents & Tool Use
τ²-bench · AA13.2#216
Writing & Language
IFBench (instructions)15.9#247
FAQ
How much does Granite 4.0 350M cost?›
There is no per-unit API price: self-host only.
Is Granite 4.0 350M free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Granite 4.0 350M, and what is it best at?›
#257 of 334 in Chat & Knowledge, #210 of 239 in Coding, #213 of 242 in Agents & Tool Use, #244 of 245 in Writing & Language.
Is Granite 4.0 350M better than Gemma 3n E4B Instruct?›
On AA Intelligence Index 4.3.2, Granite 4.0 350M scores 4.8 vs 4.8 — about the same.
What is the context window of Granite 4.0 350M?›
33K tokens.
When was Granite 4.0 350M released? Is there a newer version?›
Released 2025-10-28. It is the current version in its line.
Is Granite 4.0 350M open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Granite 4.0 350M locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Granite 4.0 350M read images?›
No, it accepts text only.
What is Granite 4.0 350M?›
Granite 4.0 350M is an open-weights language model from IBM, released Oct 2025.
What are the best alternatives to Granite 4.0 350M?›
GPT-6 Luna (better), K2 Horizon 7B (open weights).