Claude-Code-Optimized gpt-oss:20b (32K Context)

This is a customized version of gpt-oss:20b fine-tuned for Claude-like coding behavior when used with Claude Code, Ollama’s Anthropic-compatible API, or any tool expecting Messages API + tool calling.

Key Optimizations

• 32K context window (num_ctx 32768) – perfect for large codebases, long refactoring sessions, or multi-file projects
• Low temperature (0.15–0.2) + tuned sampling (top_p 0.9, top_k 40, repeat_penalty 1.15) → clean, deterministic, low-hallucination code output
• Claude-inspired system prompt – encourages step-by-step reasoning, safety, best practices, clear explanations, and honest limitations
• Native tool calling support (no extra config needed – works out-of-the-box with tools parameter in API calls)
• MXFP4 quantization (~14 GB) – efficient on consumer hardware with good VRAM

Ideal Use Cases

• Local Claude Code replacement / companion
• Complex refactoring & clean code tasks
• Algorithm implementation with explanations & tests
• Teaching / explaining code step-by-step
• Projects needing reliable, low-creativity code generation

How to Use

”`bash

Pull the model

ollama pull slekrem/gpt-oss-claude-code-32k:20b

Run interactively

ollama run slekrem/gpt-oss-claude-code-32k:20b

For Claude Code or Anthropic-compatible clients

export ANTHROPIC_BASE_URL=http://localhost:11434/v1 export ANTHROPIC_API_KEY=ollama # dummy key – Ollama ignores it claude –model slekrem/gpt-oss-claude-code-32k:20b “Refactor this messy function…”