Verified models for code generation, completion, analysis, and review.
This category contains the best code-specialized language models, from general-purpose code assistants to models trained on hundreds of programming languages. All entries include links to weights, documentation, and benchmarks.
| Field | Details |
|---|---|
| Description | Microsoft's efficient small language model optimized for code tasks |
| Parameters | 3.8B |
| Context Length | 4K |
| License | MIT |
| Type | Completion |
| Tags | code efficient small microsoft |
| Languages | Python, JavaScript, TypeScript, Java, C++ |
| GitHub | huggingface.co/microsoft/Phi-3-mini-4k-instruct |
| Homepage | azure.com/phi |
Best for: Efficient code generation, edge deployment, resource-constrained environments.
| Field | Details |
|---|---|
| Description | Open-source code model with strong performance across 338 programming languages |
| Parameters | 16B, 236B |
| Context Length | 128K |
| License | MIT |
| Type | Completion |
| Tags | code multilingual open-source |
| Languages | Python, JavaScript, TypeScript, Java, C++ |
| GitHub | huggingface.co/deepseek-ai |
| Homepage | deepseek.com |
| Stars | 20,000 |
Best for: Multi-language code generation, long-context code analysis.
| Field | Details |
|---|---|
| Description | Code generation model trained on 619 programming languages |
| Parameters | 3B, 7B, 15B |
| Context Length | 16K |
| License | OpenRAIL-M |
| Type | Completion |
| Tags | code multilingual open-source |
| Languages | Python, JavaScript, Java, Go, Rust |
| GitHub | huggingface.co/bigcode/starcoder2 |
| Homepage | huggingface.co/bigcode |
| Stars | 10,000 |
Best for: Wide language support, code completion, fill-in-the-middle.
| Field | Details |
|---|---|
| Description | Qwen's code-specialized model with strong coding performance |
| Parameters | 7B |
| Context Length | 32K |
| License | Apache-2.0 |
| Type | Completion |
| Tags | code specialized qwen |
| Languages | Python, JavaScript, TypeScript, Java, Go, C++ |
| GitHub | huggingface.co/Qwen/CodeQwen1.5-7B |
| Homepage | qwenlm.github.io |
Best for: Code-specialized tasks, instruction-following for code, multilingual coding.
| Field | Details |
|---|---|
| Description | Mistral's first ever code model designed for code generation and completion |
| Parameters | 22B |
| Context Length | 32K |
| License | Apache-2.0 |
| Type | Completion |
| Tags | code mistral generation |
| Languages | Python, JavaScript, TypeScript, Java, C++, Bash, Spanish, French |
| GitHub | huggingface.co/mistralai/Codestral-22B-v0.1 |
| Homepage | mistral.ai |
Best for: Code generation, fill-in-the-middle, multi-language support.
| Field | Details |
|---|---|
| Description | Meta's code-specialized Llama variant for code generation and understanding |
| Parameters | 7B, 13B, 34B, 70B |
| Context Length | 16K |
| License | Llama 3.1 |
| Type | Completion |
| Tags | code llama infilling |
| Languages | Python, JavaScript, TypeScript, Java |
| GitHub | huggingface.co/codellama |
| Homepage | llama.meta.com |
| Stars | 15,000 |
Best for: Python-heavy workflows, code infilling, instruction-following.
| Field | Details |
|---|---|
| Description | Google's lightweight code model built from Gemma research |
| Parameters | 2B, 7B |
| Context Length | 8K |
| License | Gemma |
| Type | Completion |
| Tags | code lightweight google |
| Languages | Python, JavaScript, TypeScript |
| GitHub | huggingface.co/google/codegemma-7b-it |
| Homepage | ai.google.dev/gemma |
| Stars | 5,000 |
Best for: Lightweight code completion, edge deployment.
| Field | Details |
|---|---|
| Description | Alibaba's code-specialized model with strong coding performance |
| Parameters | 0.5B to 32B |
| Context Length | 32K |
| License | Apache-2.0 |
| Type | Instruct |
| Tags | code instruct multilingual |
| Languages | Python, JavaScript, TypeScript, Java, Go, Rust |
| GitHub | huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct |
| Homepage | qwenlm.github.io |
| Stars | 12,000 |
Best for: Instruct-based coding, multilingual code generation.
| Model | Parameters | Context | Languages | License |
|---|---|---|---|---|
| Phi-3 Mini | 3.8B | 4K | 5+ | MIT |
| DeepSeek Coder V2 | 16B, 236B | 128K | 338 | MIT |
| StarCoder 2 | 3B-15B | 16K | 619 | OpenRAIL-M |
| CodeQwen1.5 | 7B | 32K | 6+ | Apache-2.0 |
| Codestral | 22B | 32K | 8+ | Apache-2.0 |
| CodeLlama | 7B-70B | 16K | 4+ | Llama 3.1 |
| CodeGemma | 2B, 7B | 8K | 3 | Gemma |
| Qwen 2.5 Coder | 0.5B-32B | 32K | 6+ | Apache-2.0 |
- Browse
coding-models.jsonfor all available models - Filter by parameter count, license, or supported languages
- Follow HuggingFace links for model weights and usage examples
- See
../local-inference-tools/for running models locally - See
../developer-tools/for IDE integrations
- AI Models - General language models
- Developer Tools - IDEs and code assistants
- Coding Agent Configs - Agent configurations
- Benchmarks - Performance comparisons
Status: Actively maintained | Last updated: 2026-08-17