OpenCoder 8B
Open code model trained on 2.5T tokens of permissively-licensed code — reproducibility-focused.
About
OpenCoder-8B-Instruct (November 2024) is notable for how it was built: trained on 2.5T tokens of permissively-licensed code only (the OpenCoder dataset), with the full training pipeline published for reproducibility. The authors claimed performance competitive with GPT-4 Turbo on some coding evaluations (provider-claimed).
Licensing: the Hugging Face checkpoint carries a custom "other" tag — not a standard OSI license. Verify terms before commercial use; the permissive-data story doesn't automatically mean permissive weights. 10.9k Hugging Face downloads. A 2024 8B model superseded by Qwen2.5-Coder-7B, but the reproducibility angle (open data + open pipeline) is genuinely distinctive for research.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed (paper): competitive with GPT-4 Turbo on some coding evals — treat as provider-claimed; evaluated on the authors' benchmark choices. No independent verification found.
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). The permissive-data + reproducible-pipeline story is genuinely distinctive for research. Docked five points: (1) custom checkpoint license — not OSI, verify commercial terms; (2) GPT-4-Turbo-competitive claims are provider-reported on chosen evals; (3) 2024-era 8B, superseded by Qwen2.5-Coder-7B; (4) 8K context; (5) small team, limited follow-through. For reproducibility researchers, not product builders.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.2/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 4.6/10
- Manual review · sanity read + red-flag check
- 5.0/10
- · HF downloads 10,852 (peer-normalized)
- · released 698 days ago
- · manual review: reproducibility angle distinctive; GPT-4-Turbo claims provider-reported; superseded by Qwen2.5-Coder-7B; license: custom checkpoint license (other tag) — verify commercial terms; permissive-data story doesn't mean permissive weights