CodeGen Models by Salesforce AI Research
Last refreshed 2026-10-09. Next refresh: weekly.
About
CodeGen is Salesforce AI Research’s family of autoregressive language models for program synthesis: generating code from English prompts written as comments, or completing partially written code. The first release, in March 2022, came in four sizes (350M, 2B, 6B, and 16B) and three training-data variants: NL, Multi, and Mono, where Mono models are further pretrained on Python. Salesforce followed with CodeGen2 in May 2023, which added infill sampling, and the 7B CodeGen2.5 in July 2023.
Decision facts
- Best fit
- codingcode
- Capability starting point
- CodeGen2.5 7B Mono with 2k context
- Lowest tracked input
- Not tracked
- Closest related family
- SWE-1
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs code, 2k context, and 7B parameters.
Use when the workload needs code, 2k context, and 7B parameters.
Use when the workload needs code, 2k context, and 16B parameters.
Use when the workload needs code, 2k context, and 16B parameters.
Use when the workload needs code, 2k context, and 16B parameters.
Use when the workload needs code, 2k context, and 16B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| CodeGen2.5 7B Multi | Use when the workload needs code, 2k context, and 7B parameters. | 2023-07 | code2k context7B parameters | Current |
| CodeGen2.5 7B Mono | Use when the workload needs code, 2k context, and 7B parameters. | 2023-07 | code2k context7B parameters | Current |
| CodeGen2 16B | Use when the workload needs code, 2k context, and 16B parameters. | 2023-05 | code2k context16B parameters | Current |
| CodeGen-NL 16B | Use when the workload needs code, 2k context, and 16B parameters. | 2022-03 | code2k context16B parameters | Current |
| CodeGen-Multi 16B | Use when the workload needs code, 2k context, and 16B parameters. | 2022-03 | code2k context16B parameters | Current |
| CodeGen-Mono 16B | Use when the workload needs code, 2k context, and 16B parameters. | 2022-03 | code2k context16B parameters | Current |
Release Timeline
3 release groupsSpecifications(6 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| CodeGen2.5 7B Multi | 2023-07 | 2k | 7B |
| CodeGen2.5 7B Mono | 2023-07 | 2k | 7B |
| CodeGen2 16B | 2023-05 | 2k | 16B |
| CodeGen-NL 16B | 2022-03 | 2k | 16B |
| CodeGen-Multi 16B | 2022-03 | 2k | 16B |
| CodeGen-Mono 16B | 2022-03 | 2k | 16B |
