# Best local LLM for coding — September 2026

> Best local LLM for coding (September 2026): on BenchLM's evidence, GLM-5.3 has the highest coding score estimate (61.4) among models that meet the page's constraints, followed by GLM-5.2 and Qwen3.8 Max. Evidence, constraints, and caveats included.

## Constraints on this page

A builder choosing by accuracy, running on their own hardware, ordinary input size. Every refine link below changes one answer.

## The shortlist for local LLM for coding

| # | Model | Creator | Coding score estimate | Why |
|---|---|---|---|---|
| 1 | [GLM-5.3](/models/glm-5-3) | Z.AI | 61.4 | Highest coding estimate among models that meet every stated constraint; 1M context; Open weights, so it can run on your own hardware |
| 2 | [GLM-5.2](/models/glm-5-2) | Z.AI | 60.8 | Estimate 60.8 on the same coding evidence; 1M context; Open weights, so it can run on your own hardware |
| 3 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 60.7 | Estimate 60.7 on the same coding evidence; 1M context; Open weights, so it can run on your own hardware |
| 4 | [Hy4 preview](/models/hy4-preview) | Tencent | 59.4 | Estimate 59.4 on the same coding evidence; 1M context; Open weights, so it can run on your own hardware |
| 5 | [Ornith-1.5-397B](/models/ornith-1-5-397b) | Ornith AI | 59.0 | Estimate 59.0 on the same coding evidence; 262K context; Open weights, so it can run on your own hardware |

Ordered by task score under the stated constraints.

## What this shortlist rests on

- 25% · [SWE-bench Pro](/benchmarks/swe-bench-pro) · Current
- 15% · [LiveCodeBench](/benchmarks/livecodebench) · Current
- 15% · [LiveCodeBench (Vals)](/benchmarks/valslivecodebench) · Current
- 10% · [cursorBench32](/benchmarks/cursorbench) · Stale
- 10% · [SciCode](/benchmarks/scicode) · Refreshing
- 10% · [SWE-bench Verified](/benchmarks/swe-bench-verified) · Refreshing
- 10% · [SWE-Rebench](/benchmarks/swe-rebench) · Current
- 5% · [SWE Multilingual](/benchmarks/swe-bench-multilingual) · Current

## What to verify before choosing

- Open weights are only the first requirement. Hardware, quantization, license, and tool access need verification.
- Composite scores are estimates. A small score gap does not establish a reliably better model.

## Refine

- [Change any answer in the selector](/tools/llm-selector?audience=builders&family=coding&useCase=local-code&priority=accuracy&privacy=hosted&context=normal)
- [Cost matters most](/tools/llm-selector?audience=builders&family=coding&useCase=local-code&priority=cost&privacy=hosted&context=normal)
- [Speed matters most](/tools/llm-selector?audience=builders&family=coding&useCase=local-code&priority=speed&privacy=hosted&context=normal)
- [Long documents or a large codebase](/tools/llm-selector?audience=builders&family=coding&useCase=local-code&priority=accuracy&privacy=hosted&context=large)
- [Choosing an everyday assistant, not an API](/tools/llm-selector?audience=everyday&family=coding&useCase=local-code&priority=accuracy&privacy=hosted&budget=any&context=normal)
- [Describe your own job instead](/tools/llm-selector)

## Questions

**Which is the best local LLM for coding?**

On BenchLM's public evidence, GLM-5.3 by Z.AI has the highest coding score estimate among models that meet the page's default constraints (61.4). Ordered by task score under the stated constraints. Open weights are only the first requirement. Hardware, quantization, license, and tool access need verification.

**What are the alternatives to GLM-5.3 for local LLM for coding?**

GLM-5.2 (60.8), Qwen3.8 Max (60.7), Hy4 preview (59.4), Ornith-1.5-397B (59.0) follow on the same evidence. A small gap does not establish a reliably better model; compare them on three representative examples of your own work.

**How does BenchLM pick the best local llm for coding?**

The page runs the LLM Selector with fixed answers: a builder choosing by accuracy, running on their own hardware, ordinary input size. The selector uses the coding evidence surface, filters by the stated constraints, and orders by that evidence. It never adds a hidden fit score or a bonus for open weights or reasoning style.

**Can I change the constraints?**

Yes. Every link under "Refine" opens the selector with one answer changed, and the address carries the answers so a result can be shared or reopened against the current dataset.

Data updated 2026-09-18. Method: bench-align-v5.5-2026-09-04.

Canonical page: https://benchlm.ai/best/for/local-code