Claude Opus 4.7 Cost & Pricing (May 2026)
Per-1M-token rates for every Claude Opus 4.7 tier — standard, cached, batch, and stacked. Cost-per-task estimates for typical workloads and a like-for-like comparison vs GPT-5.5, Gemini 3.1 Pro, and DeepSeek V4 Pro.
Pricing tiers — every way to buy Opus 4.7
| Tier | Input /1M | Output /1M | Notes |
|---|---|---|---|
| Standard (sync) | $5.00 | $25.00 | List price for the synchronous Messages API. The published rate. |
| Cached input | $0.50 | $25.00 | Cache hits are 90% off list on input. Output rate unchanged. Cache writes are 25% more expensive than list (5-min TTL). |
| Batch (24h SLA) | $2.50 | $12.50 | 50% off list for asynchronous workloads via the Message Batches API. Stackable with prompt caching. |
| Cached + Batch | $0.25 | $12.50 | Stacked discount: cached input on the batch tier. The cheapest way to run Opus 4.7 — but only useful for repeatable async work. |
All prices in USD per 1M tokens. The Opus 4.7 tokenizer produces roughly 35% more tokens per English input than Opus 4.6, so effective bills rose ~33% on like-for-like prompts at unchanged list prices. Re-baseline before migrating.
Cost per task — what you actually pay
| Task | In tok | Out tok | Standard | Cached | Batch |
|---|---|---|---|---|---|
| Short chat reply | 800 | 200 | $0.0090 | $0.0054 | $0.0045 |
| Long-doc summary (50-page PDF) | 80,000 | 1,500 | $0.4375 | $0.0775 | $0.2188 |
| Agentic loop (12 tool turns) | 45,000 | 6,000 | $0.3750 | $0.1725 | $0.1875 |
| RAG query (10-doc context) | 12,000 | 600 | $0.0750 | $0.0210 | $0.0375 |
Cost per single invocation. Cached column assumes 100% cache hit on input — real-world hit rates of 70-90% are typical with a well-structured system prompt and tool-definition prefix.
Opus 4.7 vs nearest alternatives
| Model | In /1M | Out /1M | Context | Note |
|---|---|---|---|---|
| Claude Opus 4.7 | $5.00 | $25.00 | 1M | This page. Coding Arena #1 at 1567 Elo. |
| GPT-5.5 | $5.00 | $30.00 | 1M | Same input, +20% output. Stronger at voice and ecosystem tooling. |
| GPT-5.5 Pro | $30.00 | $180.00 | 1M | 6x the price. Marginal lift on hardest reasoning. Rarely worth it for coding. |
| Gemini 3.1 Pro | $3.50 | $10.50 | 2M | 30% cheaper input, 58% cheaper output. Better for long-context and science. |
| DeepSeek V4 Pro | $1.74 | $3.48 | 1M | ~7x cheaper output. Apache 2.0 — also self-hostable. |
When Opus 4.7 is worth the price
- Coding agents. Coding Arena #1 at 1567 Elo and SWE-bench Pro 64.3% — the gap to second place is large enough to justify the premium on engineering workloads where output quality drives downstream cost.
- Long-horizon agentic loops. Tool-use reliability and 1M-token context make 12+ turn loops practical without hand-holding.
- High-value writing and analysis. If a single output is worth more than $5, the model fee is rounding error.
When to switch to a cheaper alternative
- Sonnet 4 ($3 / $15) — covers ~70% of typical agentic and chat workloads at one-fifth the price.
- Gemini 3.1 Pro ($3.50 / $10.50) — better for long-context, multimodal, and scientific reasoning. 2M context.
- DeepSeek V4 Pro ($1.74 / $3.48) — ~7x cheaper output. Apache 2.0 also makes self-host viable for sovereignty or zero-marginal-cost agentic loops at scale.
- Haiku 3.5 ($0.80 / $4) — for classification, routing, and extraction where Opus quality is wasted.
Related
- AI Model Leaderboard — quality vs price across all providers
- Per Million Tokens True Cost — hidden adders pushing bills 1.5-3x above list
- Token Cost Calculator — interactive estimator
- Claude Opus 4.7 deep-dive — full benchmarks and architecture
Teams running Opus 4.7 alongside other providers typically front the API with Swfte Connect to route across these models behind one OpenAI-compatible surface with prompt caching and per-route fallback.
Sources: official Anthropic pricing page, May 2026-05-06. Tokenizer drift figures from Swfte Connect telemetry on representative SaaS workloads.