AI
Claude
Anthropic's frontier model family.
Claude is the model I reach for first. Opus for hard reasoning, Sonnet for production, Haiku for speed — all built around extended thinking, tool use, and the safest pairing for agentic work.
- Long-context windows up to 1M tokens for whole-repo reasoning
- Native tool use + computer use + extended thinking
- Best-in-class coding & writing for production work
Why
Most teams pick the wrong model and blame the prompt. Claude is the safest default — strongest at code, longest context, cleanest tool use — so the rest of the stack stops fighting the model.
How
- Opus on hard reasoning, Sonnet for production defaults, Haiku where speed wins
- Plan mode + skills + hooks wired in from day one
- Cost ceilings + routing rules so the bill never surprises you
Proof
- Daily POMs
- ≈10
- Apps built on Claude
- 13 live
- First-attempt PR pass rate
- 82%
Claude — three model tiers, one router
Match the task to the model
Hover or tap a node to see details.
FAQ
- Which Claude model should I use by default?
- Sonnet for production workloads — it's the workhorse, 80% of Opus capability at 20% of the cost. Reach for Opus only when correctness > speed/cost (hard reasoning, long-context refactors). Haiku for high-throughput background tasks.
- How big is the context window in practice?
- Up to 1M tokens on the right variants — enough to drop a whole repo in. In practice I rarely need more than ~250k; long context is most useful for whole-codebase reasoning, not stuffing everything into one prompt.
- Does Claude support tool use and computer use?
- Yes — both. Tool use is the foundation of every agent I ship; computer use is the lift when the agent has to drive a UI nothing exposes via API. Both are first-class in the SDK.
In production
- matsiems.com
Whole site built and maintained on Claude Sonnet — 34 leaf pages, 17 diagrams, 51 FAQs, full admin in one POM-month.
See it - Fleet PR pass rate
82% first-attempt PR pass rate across 13 production sites on Claude Sonnet + Opus routing.
- AIOS router
Sonnet handles 70% of fleet workloads, Opus the hard 20%, Haiku the high-volume 10%.
See it