TL;DR — Opus 5 runs 2.5× Sonnet 5's rate ($5/$25 vs $2/$10 — and Sonnet's $2/$10 is now permanent, not a promo). Every number below is from our weekly-verified pricing tracker (verified 2026-09-13).
Pricing side by side (per 1M tokens, USD)
| Model | Provider | Input | Output |
|---|---|---|---|
| Claude Opus 5 | Anthropic | $5 | $25 |
| Claude Sonnet 5 | Anthropic | $2 | $10 |
Verified 2026-09-13 against official rate cards — the tracker keeps the change log. Note: Claude Sonnet 5 — introductory $2/$10 made permanent; the planned Sept 1, 2026 rise to $3/$15 was cancelled.
What a real month costs
At a standard workload of 10M input + 2M output tokens/month:
- Claude Opus 5: $100.00
- Claude Sonnet 5: $40.00
On this mix, Claude Sonnet 5 is the cheaper column (2.5× gap). Your ratio will differ — run your own mix through the LLM cost calculator.
Choose Claude Opus 5 if
correctness is the budget: multi-step agent chains, code that ships, analysis someone acts on. The 2.5× premium disappears the first time it saves one bad output.
Choose Claude Sonnet 5 if
you are doing everyday reasoning in bulk. Sonnet 5 is aggressively cheap for its tier, and that rate is now permanent — you can size volume estimates against it without a expiry caveat.
Routing decision card
- Sonnet 5 as default, Opus 5 on escalation — the classic two-rung Anthropic setup
- The escalation threshold is now stable: Sonnet's $2/$10 is permanent, so you are not re-tuning it on a promo calendar
- Cache both: Anthropic's ~90% cached-input discount applies at every tier
- The router answer: most teams shouldn't hard-pick one — ClawRouters routes each request to whichever side fits it, so the comparison becomes a per-call decision instead of a bet.
Per-model detail: Claude Opus 5 deep dive
Part of the LLM model comparison index. Prices move fast in 2026 — the tracker logs every change with dates.