CalcSays
MODEL vs MODEL · AT YOUR WORKLOAD

Compare LLM API Costs

Pick any two of 315 models and price them at your workload — real monthly bills with output skew and caching, plus the output ratio where the cheaper one flips. Not a spec sheet.

For engineers choosing between two models — compares real bills at your workload (output skew + caching included) and shows where the answer flips, not a spec sheet.

Model prices from OpenRouter · updated 2026-07-13

01 Your matchup & workload

Model A
Model B
Prompt cachingapplied per model where priced

02 At your workload

Claude Sonnet 5 wins by $5,628/month (66%) here.

Claude Sonnet 5 · cheaper
$2,952
$3.51/M effective · $2/$10 listed
GPT-5.5
$8,580
$10.21/M effective · $5/$30 listed

No flip at this input size — Claude Sonnet 5 wins at every output length.

ctx 1000K · cache $0.2/M
ctx 1050K · cache $0.5/M
Tweet
📄 Get the full agent cost audit
A multi-page report on your exact setup: where cost spirals, how much context-trimming / state-resets / model-routing each saves, and which model+framework combo is cheapest at your real failure rate. Generated from your numbers.

Related cost calculators

Claude Sonnet 5 API cost →GPT-5.5 API cost →Claude Sonnet 5 agent cost →GPT-5.5 agent cost →LLM API Cost Calculator →Compare LLM Costs →

Prices from OpenRouter, snapshot 2026-07-13, synced daily. Each model billed at your workload with its own cache-read pricing on the cached share; flip point solves input × (effInA − effInB) + output × (outA − outB) = 0. Cost is a shortlist — quality on your evals decides. All math runs in your browser.

How the math works

Spec-sheet comparisons list prices; this page prices a workload. At the defaults — 2,000 input + 800 output tokens per request, 300,000 requests a month, 60% cached prefix — Claude Sonnet 5 bills $2,952/month against GPT-5.5's $8,580: 66% apart.

Claude Sonnet 5 lists $2/M in and $10/M out (cache reads $0.2/M); GPT-5.5 lists $5/M and $30/M (reads $0.5/M). Headlines don't settle it — the blend of your input/output mix and each model's caching does.

At this input size the answer never flips: one model is cheaper at every output length, so the choice is about quality and context window, not the output ratio.

Cost is the shortlist, not the verdict: context windows (Claude Sonnet 5: 1000K vs GPT-5.5: 1050K), caching support, and quality on your evals decide the rest. Prices sync daily from OpenRouter; all math runs in your browser.

Frequently asked questions

Which is cheaper, Claude Sonnet 5 or GPT-5.5?

At this page's default workload, Claude Sonnet 5: $2,952/month vs $8,580 — 66% less. But the gap depends on your input/output mix. Tune the sliders above for your real shape.

Why not just compare the advertised prices?

Because they're input rates. Claude Sonnet 5 charges $10/M for output and GPT-5.5 $30/M — and output typically drives 60–85% of a bill. Add caching ($0.2/M vs $0.5/M reads) and two same-headline models can differ 2–3× on a real workload.

When does the cheaper model flip?

At this input size, never — one model wins at every output length. The flip only exists when one model has cheaper effective input and the other cheaper output.

Is the cheaper model the right choice?

Only if quality holds on your task. Use this page to quantify the price of preference: if GPT-5.5 wins your evals, the premium is $5,628/month at these defaults — sometimes that's obviously worth it, sometimes obviously not. Context windows and caching support (both shown above) can also decide it outright.

Are these prices current?

Prices sync daily from OpenRouter's public catalog and the page shows its snapshot date. If a sync fails, the last verified snapshot keeps serving. All math runs client-side with tested code.