| title |
aliases |
tags |
sources |
created |
updated |
| Claude / Anthropic Model Catalog |
| claude-models |
| anthropic-models |
| opus-sonnet-haiku |
|
| claude |
| anthropic |
| models |
| llm |
| ai |
|
| knowledge cutoff Aug 2025 + Claude Code system context |
|
2026-04-17 |
2026-04-17 |
Model Families
| Family |
Tier |
Best for |
| Opus 4.x |
Most capable |
Complex reasoning, architecture, long-form writing |
| Sonnet 4.x |
Balanced |
Coding, agentic tasks, daily work — best cost/performance |
| Haiku 4.x |
Fast & cheap |
Logging, classification, summaries, bulk operations |
Current Model IDs (Claude 4.x)
| Model |
ID |
Context |
Notes |
| Opus 4.7 |
claude-opus-4-7 |
200K |
Most powerful. Fast Mode available (same model, faster output) |
| Sonnet 4.6 |
claude-sonnet-4-6 |
200K |
Default in Claude Code. Best everyday coding model |
| Haiku 4.5 |
claude-haiku-4-5-20251001 |
200K |
Use for logging/notes in Obsidian vault (per CLAUDE.md) |
When to Use Each Model
Opus — multi-step reasoning, large refactors, security reviews, novel architecture decisions
Sonnet — default for all coding tasks; Claude Code uses this; agentic pipelines, tool use, RAG
Haiku — Obsidian session logging, bulk text processing, classification, any "just write text" task; cheapest
Claude Code Context
- Claude Code CLI currently runs on Sonnet 4.6 by default
/fast toggles Fast Mode on Opus 4.6 (faster output, same quality)- Memory compiler
flush.py / compile.py should use Sonnet; Obsidian logging uses Haiku (per vault CLAUDE.md)
- Agent SDK: pass model ID in
ClaudeAgentOptions → model field
API Usage
from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(
model="claude-sonnet-4-6", # or opus-4-7, haiku-4-5-20251001
max_tokens=1024,
messages=[{"role": "user", "content": "Hello"}]
)
Key Takeaways
- Default to Sonnet 4.6 for agentic/coding work — best balance
- Use Haiku for any high-volume or low-stakes generation (logging, summaries) to cut costs
- Use Opus only when the task genuinely requires maximum reasoning
- Model IDs include version suffix — always use full ID in code to avoid silent upgrades
- All 4.x models support 200K context window
Related
