Granite 4.2 family
Reasoning, tool use, multilingual generation, and coding; the 4.2 8B card documents 128K context and local Transformers/vLLM serving.
Score calculation
(2 + 5 + 7 + 6 + 3) ÷ 5 = 4.6
Pressure target: −1 overall point. Overall score: 3.6/10.
5 of six dimensions rated. Unrated dimensions are excluded from the average.
Full scoring method ↗Setup details
IBM Granite 4.2 language models, including 3B, 8B, and 30B released checkpoints; later Granite families are not included.
What fits your needs
IBM Granite 4.2 language models, including 3B, 8B, and 30B released checkpoints; later Granite families are not included.
Good fit
Reasoning, tool use, multilingual generation, and coding; the 4.2 8B card documents 128K context and local Transformers/vLLM serving.
Tradeoffs to expect
Tradeoffs
Apache 2.0 Apache 2.0 weights and a public dataset/training repository support inspection and reuse. Pretraining is described at a high level and not every training data item has a rights record, limiting reproducibility.
How to switch
Local setup
The official card provides local Transformers and vLLM instructions. Check exact memory and runtime needs for the selected 3B/8B/30B checkpoint; the 30B model is not a generic laptop recommendation.
Practical questions
Weights and license
Fetch the exact checkpoint revision and read its own license; sibling and distilled variants may inherit different terms.
Hardware
The official card provides local Transformers and vLLM instructions. Check exact memory and runtime needs for the selected 3B/8B/30B checkpoint; the 30B model is not a generic laptop recommendation.
Training data
Treat the weight license and public web/data availability separately from evidence of permission for the training data.
This is a practices and disclosure score for the named release, not a request-level energy or emissions estimate.
See the evidence
IBM identifies a GB200 NVL72 cluster hosted by CoreWeave for Granite 4.2 training, but reports no measured electricity, emissions, or water for this checkpoint. The rubric rewards documented reporting and controls; it does not infer high consumption from missing measurements.
1 + 1 reporting + 0 electricity + 0 controls = 2/10. This grades documented practices and transparency.
No comparable per-request electricity, emissions, water, regional serving mix, or independently reviewed model footprint is established in the cited sources.
- Reporting1/3
- IBM identifies a GB200 NVL72 cluster hosted by CoreWeave for Granite 4.2 training, but reports no measured electricity, emissions, or water for this checkpoint.
- Electricity sourcing0/3
- No model-specific clean-electricity sourcing or hourly power evidence is published in the reviewed sources; zero means unestablished, not dirty electricity.
- Resource controls0/3
- No model-specific documented control or measured physical resource reduction is established by the reviewed release materials. Public weights or smaller variants alone do not prove reduced resource use.
Water disclosure
Water reporting has not been assessed for this choice.
The release provides some training-process information, but not a complete permission ledger for inherited and later training data.
See the evidence
IBM lists stages, a training-data repository, permissively licensed public datasets, synthetic examples, agent traces, and human-authored material. Its 4.1 inherited base and curated subsets still lack a complete rights ledger. A public weight license governs use of those weights; it is not proof that every training source was permitted.
1 base + 2 source transparency + 2 documented permission + 0 creator controls = 5/10. Weight licensing is not treated as training data permission.
Complete source lists, source-specific rights, inherited-base provenance, synthetic-data generation permissions, and creator removal or retraining remedies remain unestablished.
- Source transparency2/3
- IBM lists stages, a training-data repository, permissively licensed public datasets, synthetic examples, agent traces, and human-authored material. Its 4.1 inherited base and curated subsets still lack a complete rights ledger.
- Documented permission2/4
- Some named or permissively licensed material is described, but the sources don’t settle permission for the full inherited and later-stage training data.
- Creator controls0/2
- The reviewed model documentation doesn’t show creator permission covering all training data, opt-out, removal, or retraining remedy for this model family.
Privacy depends on where these released weights are run; the model family has no single prompt-retention or training route.
See the evidence
The official release makes weights available for separate deployments, but it does not define one inference host or retention policy for every user. Local and hosted operators may handle prompts differently, so a single numerical privacy grade would misstate this model-level entry.
The user-selected host, region, telemetry, logging, extensions, and provider training terms are outside the checkpoint release and require a route-specific review.
Apache 2.0 weights and a public dataset/training repository support inspection and reuse. Pretraining is described at a high level and not every training data item has a rights record, limiting reproducibility.
See the evidence
Apache 2.0 weights and a public dataset/training repository support inspection and reuse. Pretraining is described at a high level and not every training data item has a rights record, limiting reproducibility. This openness rating concerns artifacts and reproducibility only; it does not assess whether the original training data was authorized.
Apache 2.0 Model weights and documentation are available, but limitations described above reduce source-level reproducibility.
The checked release and license apply to the named checkpoints; sibling variants, future versions, and the underlying training data may have different terms or disclosures.
IBM publishes global employee standards, workforce indicators, employee reporting channels, and supplier conduct expectations; none is specific to Granite training labor.
See the evidence
IBM reports pay-equity and workforce-safety indicators and sets standards for wages, hours, freedom of association, and suppliers. It also describes employee and supplier reporting channels with non-retaliation. These group-level protections don’t confirm conditions for Granite data, evaluation, or infrastructure workers.
1 base + 2 conditions + 2 voice and remedy + 1 assurance = 6/10. IBM documents group-level systems; evidence for Granite labor and independent outcomes is limited.
The sources do not confirm which Granite contributors or suppliers are covered, or publish AI-workforce-specific pay, locations, grievance outcomes, or independent labor-audit results.
- Pay & conditions2/3
- IBM publishes global employee standards on wages, hours, freedom of association, safety and supplier expectations, plus aggregate workforce metrics; it does not show model-specific coverage.
- Worker voice & remedy2/3
- IBM describes Open Door/Speak-Up and external channels, review of reports, and non-retaliation; no Granite-specific grievance data is disclosed.
- Evidence & follow-through1/3
- IBM describes compliance reviews and supplier standards, but these sources do not provide independent worker-level assurance or remediation outcomes for Granite labor.
IBM publicly markets defense AI and was selected for the Missile Defense Agency SHIELD contract; neither source establishes Granite deployment.
See the evidence
IBM says it will support the MDA SHIELD program and promotes AI for defense and intelligence work. Granite weights use Apache 2.0 terms without a military-use prohibition in the reviewed card; no source links this checkpoint to that contract or describes its military end-use oversight.
1 + 0 use restrictions + 0 enforcement & oversight + 2 relationship disclosure = 3/10. Known defense ties lower this rating; safeguards can improve it within the relationship’s cap.
No Granite-specific defense contract, deployment, military-use restriction, or downstream enforcement record was established.
- Use restrictions0/4
- The Granite card lists Apache 2.0 and no model-specific military-use restriction; the license is not a downstream deployment control.
- Enforcement & oversight0/3
- IBM describes governance for defense customers, but the sources don’t show monitoring or enforcement over use of downloadable Granite weights.
- Relationship disclosure2/2
- IBM publicly identifies its SHIELD selection and markets AI-enabled defense and intelligence solutions.
IBM publicly supports the MDA SHIELD program and sells AI-enabled defense/intelligence services. This is a company relationship; no Granite checkpoint deployment is established. The relationship limits this grade to 3/10.
Military rubric ↗IBM is named in the BNC No Tech campaign’s discussion of technology companies; this is a scoped pressure-campaign match, not a universal BDS status.
See the evidence
The official BNC No Tech campaign page names IBM alongside other technology companies in its discussion of systems serving the Israeli military. This supports a pressure-campaign match at the IBM parent-company level; it doesn’t confirm a use or contract involving every Granite checkpoint.
This match is limited to the named companies in the BNC No Tech campaign. It does not cover every BDS campaign, local group, investment, supplier, or product deployment; lists and ownership can change.
All sources
- Official model card or model release repository ↗
- Official source repository ↗
- Official model card or model release repository ↗
- BNC official campaign target list ↗
- BNC official campaign target list ↗
- IBM Data and Policies ↗
- IBM Risk, Compliance and Integrity ↗
- IBM to Support Missile Defense Agency SHIELD Contract ↗
- IBM Defense and Intelligence Solutions ↗