Replit Code v1.5
Replit's open 3.3B code completion model, trained on 1T tokens of permissively licensed code.
About
Replit Code v1.5 is a 3.3B-parameter causal language model from Replit focused squarely on code completion. It was trained in bfloat16 on 1 trillion tokens — roughly 200B tokens over five epochs — drawn from permissively licensed code in BigCode's Stack Dedup, filtered Markdown and reStructuredText, and a developer-oriented StackExchange sample, covering 30 programming languages. Training ran on 128 H100-80GB GPUs on the MosaicML platform with a custom 32K-token GPTNeoX vocabulary tuned for code compression.
It is Apache 2.0 licensed with no commercial-use restrictions, which made it a popular base for fine-tuning and IDE autocomplete plugins in late 2023. The 4K context window and small size are modest by today's standards, but it remains a clean, permissively licensed completion model for lightweight local setups.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published HumanEval-style benchmark table found in official materials; positioned as a completion model rather than a benchmark competitor
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: official Hugging Face model card with full training details (1T tokens, 128 H100s, 30 languages) and Apache-2.0 license; released by Replit as a commercial-grade open model. Strengths: clean permissive licensing, transparent data sourcing, widely reused as a fine-tuning base. Docked for: (1) no published benchmark scores to compare against; (2) no maintenance since 2023 and superseded by stronger open completion models.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Full training details published; popular fine-tuning base in 2023
- · No published benchmark scores; unmaintained since 2023