Anthropic Launches Claude Opus 4.6 Fast Mode at $30/$150 Per Million — 6x the Standard Price
Anthropic has added a fast-mode variant of Claude Opus 4.6, available now via API and OpenRouter. The capabilities are identical to the standard Opus 4.6 — the difference is entirely speed and price.
The Numbers
| Variant | Input | Output | Throughput | TTFT |
|---|---|---|---|---|
| Claude Opus 4.6 | $5/M | $25/M | ~30 tok/s | ~2.5s |
| Claude Opus 4.6 Fast | $30/M | $150/M | 92 tok/s | 0.93s |
The throughput delta is roughly 3x; the price delta is 6x. Cache reads on Fast Mode price at $3/M; cache writes at $37.50/M. Context window stays at 1M tokens, max output at 128K.
What It’s For
Anthropic’s documentation frames Fast Mode around latency-sensitive deployments: real-time voice pipelines, high-turn interactive agents, and user-facing applications where response lag is a product problem. For background processing, batch jobs, or workflows that run off the critical path, the cost premium is hard to justify.
The 6:1 price ratio suggests this is not intended as the default path. It’s an escape valve for the segment of production traffic where Opus-level capability and sub-second first-token latency must coexist.
Context
Anthropic’s standard Opus 4.6 is already the top composite performer on the Stack Futures ticker at 80.8% SWE-bench Verified and 99.3% tau-bench telecom. The Fast variant doesn’t change those figures — it just changes when you can afford to wait.
At $150/M output, Claude Opus 4.6 Fast Mode is the most expensive production API on the market. OpenAI’s GPT-5.4 (xhigh) runs cheaper per token for equivalent task scores. Whether that gap is worth paying depends entirely on latency requirements.
The model ID on OpenRouter is anthropic/claude-opus-4.6-fast.