Local coding models
Coder-specialised open-weight models that run on your own hardware — the Qwen2.5-Coder family, DeepSeek-Coder, StarCoder2, Yi-Coder, CodeLlama. Filtered by HumanEval+ and MBPP+ scores where we've benchmarked.
Coding-tuned LLMs aren't just smaller versions of general-purpose chat models — they're trained on much more code, often with fill-in-the-middle objectives, and they post-train with execution feedback. The result is dramatic per-parameter strength on coding benchmarks compared to general models at the same size.
Headline benchmark on this laptop (RTX 3080, 16GB): qwen-2.5-coder-7b-instruct scored 81.1 HumanEval+ pass@1 + 66.9 MBPP+ pass@1 — comparable to commercial models 5-10x its size. The pattern repeats across the coder family: smaller models hit much higher coding scores than their general-chat siblings.
Each row links to the model's full operator notes including the actual prompting kit, recommended quantization, and benchmark scores (HumanEval+, MBPP+) we've run. Filter by 'commercial OK' if the license matters.
DeepSeek-based
DeepSeek's April 2026 frontier flagship. 1.6T total / 49B active MoE with hybrid Compressed Sparse Attention + Heavily Compressed Attention. 1M context window. Closes most of the gap with Claude Opus 4.6 on coding while
- License
- MIT · OK
- Context
- 1024K
The cost-efficient sibling of V4-Pro. 284B total / 13B active MoE, same hybrid CSA+HCA attention, same 1M context. The published FP4 + FP8 checkpoint is 160 GB, which needs 3× A100 80GB for weights alone. The mlx-communi
- License
- MIT · OK
- Context
- 1024K
DeepSeek V2.5 — merged V2 chat + Coder. Pre-V3 baseline; 21B active MoE.
- License
- DeepSeek License · OK
- Context
- 128K
Full DeepSeek Coder V2. 236B total / 21B active MoE coder.
- License
- DeepSeek License · OK
- Context
- 128K
R1 reasoning distilled into a Llama 3 8B base. Smaller R1 distill; useful when 32B is too heavy. Reasoning quality is meaningfully below the 32B distill but still beats non-reasoning Llama 8B on math/code.
- License
- MIT · OK
- Context
- 128K
Qwen-based
Qwen 3.6 27B dense (not MoE) with Multi-Token Prediction. Sits between the 14B and 35B-A3B as a "single dense model with MTP throughput acceleration." Targets workloads where the MoE activated-param dance isn't ideal but
- License
- Apache-2.0 · OK
- Context
- 256K
Smallest Qwen 2.5 Coder. Targets edge / autocomplete on integrated GPUs and Apple Silicon laptops.
- License
- Apache 2.0 · OK
- Context
- 32K
Compact Qwen 2.5 Coder. Sweet spot for laptop autocomplete and small refactor agents.
- License
- Qwen Research Li
- Context
- 32K
Coding-specialized Qwen 2.5 at 14B. The 16GB-VRAM tier coding model — fits comfortably with 8K context.
- License
- Apache 2.0 · OK
- Context
- 128K
CodeQwen 1.5 — Qwen Coder predecessor. Superseded by Qwen 2.5 Coder for new deployments.
- License
- Tongyi Qianwen L · OK
- Context
- 64K
Other / from-scratch
InclusionAI's Ring-2.6-1T is a 1 trillion parameter Mixture-of-Experts model with 8 of 256 routed experts active per token, released by Ant Group's AI research arm. Targets frontier reasoning and code at MoE serving cost
- License
- MIT · OK
- Context
- 125K
StarCoder 2 flagship. The largest BigCode coder; 16k context with strong fill-in-middle.
- License
- BigCode OpenRAIL · OK
- Context
- 16K
BigCode's StarCoder 2 at 3B. Trained on 17 programming languages from The Stack v2.
- License
- BigCode OpenRAIL · OK
- Context
- 16K
Mid-size StarCoder 2. The 8GB-VRAM autocomplete pick.
- License
- BigCode OpenRAIL · OK
- Context
- 16K
Yi-based
opencoder
Llama-based
Mistral-based
Coding agent setup?
Cross-reference the model rows here with the runtime guidance at /apps for Cline, Aider, Continue, OpenInterpreter, and Claude Code adapters. The 'best coding agent for local models' Q&A digs into the actual workflows. Read the coding-agent comparison.