Why Sonnet, not Opus — deeper.

Sonnet is Anthropic's mid-tier flagship. Opus is the largest and most capable. When people ask why LADLE runs on Sonnet, the fast answer is "cost." The full answer is more interesting.

The pricing gap

Sonnet is $3/MTok input, $15/MTok output. Opus is $5/MTok, $25/MTok — roughly 1.7-2× more expensive per token. But actual model behavior differs by more than that. Opus tends to write longer outputs, use more thinking tokens, and take more turns to arrive at answers. Real-world cost per equivalent task is often 3-5× Sonnet.

At LADLE's average subscriber usage (~800K tokens/month), Sonnet inference costs us ~$6-7. Opus inference would cost ~$25-35. Building a $20/mo product around $30 of variable cost isn't a product; it's a subsidy that runs out.

Where Opus actually pulls ahead

Anthropic's own model benchmarks (which we read alongside independent evaluations) show Opus opening up a real gap on: - **Multi-step reasoning on hard math or physics problems.** - **Long agentic workflows** chaining many tool calls across many turns. - **Some code tasks on very large codebases** where the model needs to hold dozens of files in working memory. - **Complex research analysis** requiring careful cross-referencing.

For those tasks, Opus is meaningfully better. If you do them daily, the right recommendation is Claude Max ($100-200/mo directly through Anthropic) — it's the honest tool for that job.

Where the gap is small

For everyday assistant work — email drafting, document summarization, code review on typical PRs, writing help, meeting notes, research assistance, brainstorming, most everyday questions — Sonnet is close enough to Opus that most users can't tell the difference in blind tests. Anthropic's own benchmarks show this: the delta shows on hard reasoning, not on everyday assistance.

The design bias

LADLE's product opinion is that the model should be quiet and helpful, not the smartest possible entity you have access to. The frontier-model race is a fun engineering story. It isn't what we're trying to sell.

We're selling a competent daily assistant, powered by a well-known model, with a specific and unusual social contract. Sonnet is exactly the right tier for that pitch. Opus at $20/mo would break the meal-donation math and either compromise the meals or lose money.

The cost path if Anthropic changes pricing

Anthropic has cut Sonnet prices twice since LADLE launched. If they ship a Sonnet successor at meaningfully lower price, we pass the room through as either more per-subscriber headroom or additional margin (the margin pass 2026-08 chose margin — revisit with real data).

If they ever ship an Opus-tier model at Sonnet-tier prices, we switch. Until then, Sonnet is the answer and it's the honest answer.

What we don't tell users

LADLE's marketing doesn't say "we use the second-best model to save money." That's true but not the frame. The frame is: we use a model that's genuinely good enough for the tasks people actually do, and the alternative wouldn't fit our economics. Both framings are honest; the second one is more accurate to how we think about the choice.

When you should use Opus directly

If your work involves: - Regular multi-hour agentic coding sessions. - Research on hard technical problems where correctness at scale matters. - Long-horizon reasoning chains.

...Opus through Claude Max ($100-200/mo direct with Anthropic) is the right tool. LADLE isn't trying to be that product.