Granite 4.2 30B Benchmarks, Specifications & Availability

Granite 4.2 30B

Explore Granite 4.2 30B from IBM: published specifications, source-linked vendor benchmarks, independent evaluator coverage and recorded pricing when available.

Published specifications

Provider: IBM
Access: Open weights
License: Apache-2.0
Context window: 128K
Total parameters: 30B
Active parameters: Not published
Released: 2026-08-25
Modalities: text
Family: Granite 4.2

Hugging Face · Announcement

Model notes

IBM’s Granite 4.2 dense reasoning release of 25 August 2026 (card and IBM’s Hugging Face blog), post-trained from Granite-4.1-30B-Base with built-in thinking (full thinking by default, non-thinking and low-effort modes). Official Apache-2.0 weights on Hugging Face, ungated (revision 9e668ce1c538); safetensors total 29,276,770,304 parameters against the card’s “30B”. Native context 128K; the card also states long-context extension to 512K, which is not the native window. Text in and out. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK and names no thinking mode or effort for its table, so scores carry effort unknown. The card’s recommended sampling (temperature 1.0, top_p 0.95) is inference guidance, not a stated evaluation setting. No OpenRouter listing at this check, so no price row. Not to be confused with Granite 4.1 or 4.0.

Compare Granite 4.2 30B with other models →Compare subscription plans Explore data coverage

Pricing · OpenRouter

No OpenRouter pricing is recorded for this model. Missing rates are not free.

Official / vendor benchmarks

Default headline records. Own-vendor, peer-vendor and third-party provenance remain visible in evidence; configurations may differ.

Official / vendor headline scores; expand evidence for every record
Benchmark / evaluator	Headline score	Evidence
SWE-bench Multilingual	41.9%unknown effort	The recorded result & source 41.9% · raw 41.89 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “SWE Bench Multilingual”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Agent harness unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
SWE-bench Pro (revision unspecified)	33.3%unknown effort	The recorded result & source 33.3% · raw 33.29 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “SWE Bench Pro”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Agent harness and SWE-bench Pro revision unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
SWE-bench Verified	57%unknown effort	The recorded result & source 57% · raw 57 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “SWE Bench Verified”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Agent harness unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
Terminal-Bench 2.1	29.2%unknown effort	The recorded result & source 29.2% · raw 29.24 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “Terminal-Bench 2.1”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Agent harness unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
τ³-bench (overall)	62%unknown effort	The recorded result & source 62% · raw 62 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “τ³-bench (AVG)”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Average over τ³-bench domains; domain mix unstated. A different τ³-bench row (50.99 / 66.34 / 68.05) is commented out of the card source and not displayed, so it is not stored. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
BFCL v4 (overall)	61.4%unknown effort	The recorded result & source 61.4% · raw 61.39 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “BFCL (v4)”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Overall score as labelled. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
ProfBench (search setting unstated)	42.9%unknown effort	The recorded result & source 42.9% · raw 42.9 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “ProfBench”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Search setting unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
BIRD text-to-SQL (IBM Granite card; split unstated)	41.9%unknown effort	The recorded result & source 41.9% · raw 41.85 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “BirdBench”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. BIRD text-to-SQL; split and metric unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
GDPval (IBM Granite card score)	1225 scoreunknown effort	The recorded result & source 1225 score · raw 1225 score Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “GDPval”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Elo-like score; judge, reference deliverables and harness unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
AIME 2025	89.2%unknown effort	The recorded result & source 89.2% · raw 89.17 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “AIME25”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
HMMT February 2025	89.2%unknown effort	The recorded result & source 89.2% · raw 89.17 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “HMMT Feb25”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
GPQA (IBM Granite card; subset unstated)	66.4%unknown effort	The recorded result & source 66.4% · raw 66.41 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “GPQA”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. GPQA subset unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
LiveCodeBench v6	75.8%unknown effort	The recorded result & source 75.8% · raw 75.77 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “LiveCodeBench v6”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
SciCode (IBM Granite card; aggregation unstated)	38.8%unknown effort	The recorded result & source 38.8% · raw 38.76 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “SciCode”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Subproblem or main-problem aggregation unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
MMLU-Pro	77.6%unknown effort	The recorded result & source 77.6% · raw 77.6 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “MMLU-Pro”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
MMLU-ProX Lite (IBM)	66.6%unknown effort	The recorded result & source 66.6% · raw 66.64 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “MMLU-ProX lite (IBM)”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Languages unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
Arena-Hard v2 (IBM Granite card)	67.9%unknown effort	The recorded result & source 67.9% · raw 67.93 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “Arena-Hard-V2”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Judge, baseline and style control unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
IFBench	77.2%unknown effort	The recorded result & source 77.2% · raw 77.17 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “IFBench (prompt)”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Prompt-level accuracy; strict or loose unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
RULER 64K (IBM Granite card)	90%unknown effort	The recorded result & source 90% · raw 89.96 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “RULER 64K”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b
RULER 128K (IBM Granite card)	81.4%unknown effort	The recorded result & source 81.4% · raw 81.38 % Headline · unknown effort · Own vendor Source/record date: 2026-08-25 IBM Granite 4.2 30B card (revision 9e668ce1c538), evaluation table row “RULER 128K”, column “30B Dense”. IBM evaluates with a framework based on NVIDIA’s NeMo Evaluator SDK; thinking mode, effort, sampling and repeats for this table are unstated. Execution date unpublished. https://huggingface.co/ibm-granite/granite-4.2-30b

Independent evaluators

Evaluator harnesses are distinct from vendor measurements. Missing coverage is not a failed test.

No independent evaluator results are recorded for this model.

Read how we select and source scores or the comparison guide.