Can a local LLM replace Claude Code? The benchmark everyone quotes is the wrong one.

说明与用途

本地 LLM 在正确选择模型族（Qwen 3.x、GLM‑4.x、Devstral Small 2）并配置大上下文窗口后，可可靠执行编码代理任务；单个 24GB GPU 运行 Qwen 3.6 27B 是大多数用户的最佳选择。

问题

Local LLMs claim to match Claude Code on SWE‑bench, but the benchmark does not measure the reliability needed for agentic coding workflows.

结果

Local LLMs can reliably act as coding agents when the right model family (Qwen 3.x, GLM‑4.x, Devstral Small 2) is chosen and the environment is configured for large context windows; a single 24GB GPU running Qwen 3.6 27B is the sweet spot for most users.

工作流

• 1. Pull a capable coder model (e.g., qwen3-coder:30b) via Ollama.
• 2. Set environment variables to point Claude Code to the local endpoint: ANTHROPIC_BASE_URL="http://localhost:11434" and ANTHROPIC_AUTH_TOKEN="local-dummy".
• 3. Run Claude Code with the local model.
• 4. For Codex CLI, use the --oss flag to point to the local Ollama endpoint or configure LM Studio’s config.toml.
• 5. Explicitly set a large context window (32K–64K) to avoid early starvation.
• 6. Verify tool‑call reliability by testing multi‑step agentic scenarios.
• 7. Adjust GPU choice based on hardware tier: 24GB for Qwen 3.6 27B, 8GB for GLM‑4.7‑Flash, etc.

证据

SWE‑bench measures "given a bug, produce a correct patch." That's a *writing* task.

来源

Tool‑call validity — can it emit a well‑formed edit_file / run_command call the harness can parse, *reliably*, not usually?

来源

The curve climbs steeply up to ~24GB and then **flattens** — the 480B coder even dips *below* the 27B on Verified while costing 3× the memory.

来源