Model components / Fine-tuned / chat models

Granite Code Instruct

IBM's enterprise-focused open source code model line, now folded into the unified Granite 4.1 family (3B, 8B, 30B dense, base + instruct, Apache-2.0) released April 29, 2026. Granite 4.1 delivers significant improvements over Granite 4.0 in tool calling, instruction following, coding, and math, while preserving IBM's IP-clean training-data provenance, still the go-to choice for enterprises needing legal clarity on training data. Available on watsonx and Hugging Face.

IBM Granite Code Instruct family (3B/8B/20B/34B, 2K & 128K context), fine-tuned from Granite-Code-Base for code instruction-following; released May 6 2024 (arXiv 2405.04324). Instruct = post-trained (SFT on permissively-licensed code/math/NL instruction data). Verified live on HF June 2026 but marked DEPRECATED by IBM (superseded by Granite 4.x mainline).

Openness

3 high confidence

3.0

weights: open(Apache-2.0,on HF)
code: partial(inference + fine-tune sample scripts via Dolomite Engine
data: described_not_released(CommitPackFT/MathInstruct/Glaive/HelpSteer/Open-Platypus named in card but processed corpus not redistributable)
post-training-data: described(SFT mixture named, not released)
paper: open(arXiv 2405.04324)
model_card: open
license: Apache-2.0(OSI)

OSI Apache-2.0 + downloadable weights + paper + model card, but training/post-training data only described (not redistributable) and no full training pipeline => open_weights, not open_source. Strong within the open_weights class (genuinely permissive, no use caps).

https://huggingface.co/ibm-granite/granite-3b-code-instruct-2k recorded 2026-06-04
Apache-2.0 license, instruction-tuned model card, named training-data sources, HumanEvalSynthesis scores
https://github.com/ibm-granite/granite-code-models recorded 2026-06-04
repo with Apache-2.0, fine-tune sample scripts (Dolomite Engine), but no training data download and no full training pipeline
https://research.ibm.com/blog/granite-code-models-open-source recorded 2026-06-04
IBM describes data sources (GitHub Code Clean, StarCoderData, CodeNet) but does not release the processed datasets

Adoption

1 medium confidence

1.0

HF shows ~1,925 downloads/month on the granite-3b-code-instruct-2k variant; model is officially deprecated and superseded by Granite 4.x. Even summed across the four Instruct sizes the active monthly pull volume is small. Hobbyist/legacy-research tier.

https://huggingface.co/ibm-granite/granite-3b-code-instruct-2k recorded 2026-06-04
~1,925 downloads last month; deprecation notice

Capability

2 high confidence

2.0

Strong-for-size 2024 code SLMs (matched open code LLMs of their era), but sub-13B and ~51% HumanEval Python is far below 2026 frontier chat/agentic coders (Opus 4.8 88.6 SWE-bench). Low-mid on a frontier-relative scale.

https://huggingface.co/ibm-granite/granite-3b-code-instruct-2k recorded 2026-06-04
HumanEvalSynthesis pass@1 per-language scores
https://research.ibm.com/blog/granite-code-models-open-source recorded 2026-06-04
Granite-8B-Code matches SOTA among open source code LLMs on HumanEvalPack at 2024 release

Unchanged since 2026-06-24 (last edited, not re-checked)