Granite 8B Code
IBM's Apache-2.0 enterprise code model — 128K context with a commercial-friendly license.
About
Granite-8B-Code-Instruct (2024, IBM) is IBM's enterprise-oriented code model: 8B parameters, 128K context, trained on 4T tokens of code-heavy data, instruction-tuned for code generation, explanation, and fixing.
Apache 2.0 licensed — commercial use allowed, IBM's enterprise-friendly posture throughout. Direct adoption is small (681 HF downloads); Granite's usage concentrates in IBM's watsonx ecosystem rather than the open community. A 2024 model superseded by Qwen2.5-Coder-7B on raw quality, but the IBM backing, 128K context, and clean licensing make it a reasonable enterprise default where vendor relationships matter. The checkpoint was updated as recently as April 2026.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed (IBM): competitive HumanEval for 8B — treat as provider-claimed. Independent standardized scores are limited.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 5/10
Scored 5/10 (metrics tier). Apache 2.0 with IBM enterprise backing and 128K context — a reasonable corporate default. Docked five points: (1) tiny open-community adoption (681 HF downloads); (2) superseded on quality by Qwen2.5-Coder-7B; (3) benchmark claims are provider-reported; (4) real usage is watsonx-concentrated, limiting independent validation; (5) 2024-era. Pick for IBM relationships, not raw capability.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 4.7/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 9.1/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · HF downloads 681 (peer-normalized)
- · released 890 days ago
- · manual review: IBM backing, 128K context, checkpoint updated Apr 2026; tiny open adoption; superseded on quality; license: Apache-2.0