Models

Claude Code supports 100+ models across 27 providers. Pick the right model for each task — from frontier reasoning to fast local inference.

Model Selection Guide

Workload	Recommended Models	Context	Notes
Complex architecture & large edits	claude-sonnet-4-7, claude-opus-4-7, gpt-5.5-pro, deepseek-v4-pro	1M tokens	Best tool calling, strong reasoning, largest context
Code review & debugging	claude-sonnet-4-7, gpt-5.5, gemini-3.1-pro	1M-2M tokens	Good instruction following, reliable output
Quick search & summarization	claude-haiku-4-5, deepseek-v4-flash, mistral-small-latest, groq/llama-4-70b	128K-200K	Fast, cost-effective, lower latency
Offline / air-gapped	ollama/llama4:70b, ollama/llama3.3	131K tokens	Fully local, no network needed
Cost-sensitive bulk	kilocode/kilo-auto/free, opencode-go/deepseek-v4-flash-free, groq/llama-4-8b	Varies	Free or minimal cost tiers available
Research & reasoning	claude-opus-4-7, gpt-5.5-thinking, gemini-3.1-pro, deepseek-v4-pro	1M-2M tokens	Deep reasoning, chain-of-thought
Vision / multimodal	claude-sonnet-4-7, gpt-5.5-pro, gemini-3.1-flash, deepseek-v4-flash	1M tokens	Image understanding, document analysis

Notable Models by Provider

Provider	Models Available
Anthropic	claude-opus-4-7 (1M ctx), claude-sonnet-4-7 (1M ctx, recommended), claude-haiku-4-5 (200K ctx, fast), claude-sonnet-4-5 (1M ctx)
OpenAI	gpt-5.5-pro (1M ctx, reasoning), gpt-5.5-thinking (1M ctx), gpt-5.5 (1M ctx), gpt-5.4-mini (fast)
Google	gemini-3.1-pro (2M ctx), gemini-3.1-flash (1M ctx, fast)
DeepSeek	deepseek-v4-pro (1M ctx, MoE), deepseek-v4-flash (1M ctx, fast MoE)
xAI	grok-4-20 (128K ctx), grok-4 (128K ctx, recommended)
Mistral	mistral-large-4 (128K ctx), mistral-small-latest (recommended), ministral-3-14b (fast)
Groq	llama-4-70b (fast), llama-4-8b (instant)
Ollama	llama4:70b (local), deepseek-v4 (local), llama3.3 (local, verified)
GitHub Copilot	gpt-5.5, gpt-4o, claude-opus-4-7, claude-sonnet-4.5, claude-haiku-4.5
OpenCode	claude-opus-4-7, claude-sonnet-4-6, gpt-5.5, gemini-3.1-pro, kimi-k2.6, glm-5.1, minimax-m2.7
KiloCode	kilo-auto/free (free tier), gpt-5.5, gpt-5.5-pro, claude-opus-4.7, kimi-k2.6, deepseek-v4-pro, mimo-v2.5-pro
NVIDIA NIM	deepseek-v4-pro, llama-3.1-405b, llama-3.1-70b, nemotron-4-340b

Model Switching

/model                          # Interactive picker (recent models at top)
/model list                     # List all available models
/model claude-sonnet-4-7        # Switch by full ID
/model sonnet                   # Alias-based switching
--model opus                    # CLI flag at startup

Model switching is instant — the next conversation turn uses the new model. Previous context is preserved.

Model Aliases

Short aliases are resolved to full model IDs: sonnet, opus, haiku, gpt5, flash, etc.

CLI Flags for Models

Flag	Description
`--model <name>`	Set model at startup (e.g., `--model sonnet` or `--model claude-opus-4-7`)
`--effort <level>`	Set reasoning effort: `low`, `medium`, `high`, `max`
`--max-turns <N>`	Limit agentic turns in non-interactive mode
`--thinking <mode>`	Thinking mode: `enabled`, `adaptive`, `disabled`
`--fallback-model <model>`	Fallback when primary model is overloaded (print mode only)
`--task-budget <tokens>`	API-side task budget in tokens
`--max-budget-usd <amount>`	Maximum spend on API calls (print mode only)

Model Capabilities

Each model declaration in providers.json includes:

Model ID and display label
Context window (maxContext) and max output tokens (maxOutput)
Tool calling type (native, none)
Vision support
Streaming mode (full, partial)
Reasoning support
System prompt support
Tags (for filtering): fast, verified, recommended, latest, free, vision, tools, native, moe, local