TL;DR — Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens — a striking 3× price cut from Opus 4's $15/$75. Cache hits bill at ~10% of input, and the Batch API halves both sides (discounts stack). The frontier tier above it is Claude Fable 5 at $10/$50. This reprices the entire top of the market: frontier reasoning that cost $0.09 per typical call on Opus 4 now costs $0.03.
Official pricing (as of August 2026)
| Input /1M | Output /1M | |
|---|---|---|
| Claude Opus 5 | $5.00 | $25.00 |
| Cache hits | ~10% of input rate | — |
| Batch API | $2.50 (-50%) | $12.50 (-50%) |
| Reference: Opus 4 | $15.00 | $75.00 |
The batch discount stacks with prompt caching — a cached batch request can cost as little as ~5% of a standard call. For agents with stable system prompts, effective input costs land far below the headline rate.
What this costs in practice
- Coding agent planning calls (20/day, ~4K in / 1K out): ~$1.35/month — frontier reasoning is no longer the budget item it was
- Heavy agentic session (500K in / 100K out): $5.00 — versus $15.00 on Opus 4
- Batch + cache pipeline: the same session as low as ~$1.50
How it compares
| Model | Input /1M | Output /1M | Note |
|---|---|---|---|
| Claude Fable 5 | $10.00 | $50.00 | New top tier |
| Claude Opus 5 | $5.00 | $25.00 | This page |
| GPT-5.5 | $5.00 | $30.00 | Comparable input, pricier output |
| Claude Sonnet 5 | $2.00 | $10.00 | Permanent as of Sept 2026 (planned rise cancelled) |
| Claude Haiku 4.5 | $1.00 | $5.00 | Details |
Where it fits in a routing strategy
Opus 5's cut changes routing thresholds: calls that were borderline 'too expensive for frontier' now clear the bar, so well-tuned routers send more traffic to the frontier tier than six months ago while still cutting total spend — the pricing-trends guide tracks this repricing wave. ClawRouters picks the tier per request so you inherit price cuts automatically.
Routing decision card
The exclusive layer: what the 3× price cut actually changes in routing behavior.
- Route here when: multi-step planning, architecture decisions, hard debugging — the 10-20% of calls that determine whether the other 80% go in the right direction
- The repricing effect: calls that were borderline "too expensive for frontier" under Opus 4's $15/$75 now clear the bar at $5/$25 — a well-tuned router sends more traffic to frontier than six months ago while total spend still falls
- Head-to-head at this price point: GPT-5.5 ($5/$30) — identical input price, Opus 5 wins on output price; Fable 5 ($10/$50) sits above for the hardest ceiling calls
- Typical pairing: Sonnet 5 or DeepSeek V4 Pro handles implementation; Opus 5 reviews and unblocks
Prices from the provider's official rate card as of August 2026 — verify at the source before committing to volume, as rates change with model launches. Full market context: 2026 AI pricing guide · routing basics.