Language modelOpen weights

Granite 4.2 3B

IBM

Released: August 25, 2026
Data date: October 3, 2026

Granite 4.2 3B follows a shorter training schedule than 8B and 30B in IBM’s new series. The compact version receives reasoning and coding training but skips their additional agent block covering software, terminal, and search tasks.

IBM releases this dense 3-billion-parameter model under Apache 2.0. Like the larger variants, it can switch between answers with and without a thinking process.

Specifications and access

Specification	Value and source
Parameters	3 billionSource
Model class	Language model for agentic workflowsSource
License	Apache 2.0Source
Access	Open weights for local deploymentSource

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

tau3-bench (Show measurement, test conditions, and source)

Source value: 45.78
Score: 45.78
Metric: tau3-bench reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

BFCL v4 (Show measurement, test conditions, and source)

Source value: 52.41
Score: 52.41
Metric: BFCL v4 reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

ProfBench (Show measurement, test conditions, and source)

Source value: 32.1
Score: 32.1
Metric: ProfBench reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

AIME25 (Show measurement, test conditions, and source)

Source value: 78.33
Score: 78.33
Metric: AIME25 reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

GPQA (Show measurement, test conditions, and source)

Source value: 54.8
Score: 54.8
Metric: GPQA reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

LCBench v6 (Show measurement, test conditions, and source)

Source value: 69.71
Score: 69.71
Metric: LCBench v6 reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

SciCode (Show measurement, test conditions, and source)

Source value: 24.11
Score: 24.11
Metric: SciCode reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

MMLU-Pro (Show measurement, test conditions, and source)

Source value: 67.84
Score: 67.84
Metric: MMLU-Pro reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

IFBench Prompt (Show measurement, test conditions, and source)

Source value: 74.33
Score: 74.33
Metric: IFBench Prompt reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

RULER 64K (Show measurement, test conditions, and source)

Source value: 67.52
Score: 67.52
Metric: RULER 64K reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

RULER 128K (Show measurement, test conditions, and source)

Source value: 55.3
Score: 55.3
Metric: RULER 128K reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: Provider model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. IBM official technical breakdown; agent/coding values and missing sampling details are not cross-source comparable.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source Provider model card

Sources and data date

Every statement links to its underlying documentation or leaderboard.

Type	Evidence and data date
Research status	Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional source	IBM Granite 4.2 announcement (retrieved October 3, 2026) · IBM Granite 4.2 announcement · Editorial description reviewed October 4, 2026
Additional source	IBM Granite 4.2 technical report (retrieved October 4, 2026) · IBM Granite 4.2 technical report · Editorial description reviewed October 4, 2026