Choosing a Model

Claude Code allows for selecting an LLM/model to use for planning and implementation.

Model Tiers

Alias	Resolves to (Anthropic API)	Best for
opus	Opus 4.7	Complex reasoning, architecture, research
sonnet	Sonnet 4.6	Daily coding — best balance of quality/cost
haiku	Haiku (latest)	Simple scripts, fast throwaway tasks
best	Same as opus	Convenience alias
opusplan	Opus (plan) → Sonnet (execute)	Hybrid: smart planning, efficient execution

How to Set the Model

Configuration is honored in this priority order:

# 1. During session (highest priority)
/model opus
/model sonnet          # interactive picker if no arg: /model

# 2. At startup
claude --model opusplan

# 3. Environment variable
export ANTHROPIC_MODEL=sonnet

# 4. Settings file (lowest priority)
# ~/.claude/settings.json
{ "model": "sonnet" }

To persist a /model choice as your new default: press d on the highlighted row in the interactive picker — it writes to your settings file.

---

The opusplan Alias — Most Interesting Option

opusplan uses Opus during plan mode for complex reasoning and architecture decisions, then automatically switches to Sonnet for code generation and execution — giving you the best of both worlds. This is the sweet spot for complex feature work.

---

Effort Levels (Orthogonal to Model)

Effort is a separate dial that controls how much the model thinks per turn, independent of which model you're on:

Level	Use when...
low	Latency-sensitive, scoped tasks
medium	Cost-sensitive work, some quality tradeoff OK
high	Intelligence-sensitive work
xhigh	Default on Opus 4.7; best for most coding + agentic tasks
max	Deep reasoning on demanding tasks — session only, can overthink

/effort high
/effort xhigh     # Opus 4.7 default

One-off override: include ultrathink anywhere in your prompt — it triggers deeper reasoning for that turn without changing the session effort level.

Extended Context (1M tokens)

Opus 4.7, Opus 4.6, and Sonnet 4.6 all support a 1M token context window. On Max, Team, and Enterprise plans, Opus is automatically upgraded to 1M context. Sonnet with 1M context requires usage credits on every plan.

/model opus[1m]
/model sonnet[1m]

Practical Decision Framework

Is the task architectural / research-heavy?
  └─ Yes → opus or opusplan
  └─ No  → Is it simple/throwaway?
              └─ Yes → haiku
              └─ No  → sonnet (the sane default for ~90% of work)

For most daily coding sessions, sonnet with xhigh effort is the practical sweet spot. Reserve opus for when you're doing design work, deep debugging, or need architectural judgment.