What are you actually paying for with Opus?
Opus 4.7 costs $5/Mtok in and $25/Mtok out. Gemini 3.1 Pro costs $2 and $12. On 1M input plus 200K output daily, that's roughly $300/month for Opus against $132/month for Gemini. So the question isn't which is cheaper; it's whether Opus's premium buys anything your workload needs.
What it buys, in our experience, is dependability on the hardest tasks. Opus 4.7 is the model we trust with long agentic coding sessions: multi-hour runs, dozens of tool calls, changes spanning many files. Its failure mode is rarer and more recoverable. It also allows 128K output tokens per response, double Gemini's 65K, which matters exactly on those big tasks.
That premium is wasted on easy work. Classification, extraction, routine drafting: a cheaper model does these fine, and paying Opus rates for them is how API bills get silly.
Where does Gemini 3.1 Pro pull ahead?
Three places. Price, obviously, at less than half of Opus on both sides.
Input breadth, decisively. Gemini takes video and audio natively along with text, images, and PDFs. Opus stops at text, images, and PDFs. Any pipeline that touches recordings, calls, or screen captures needs either Gemini or an extra transcription stage in front of Opus.
And throughput economics. If you're processing millions of documents at tolerable quality, Gemini's rate turns projects that were uneconomical at Opus prices into routine batch jobs.
One asterisk: the preview label. Google adjusts preview pricing and behavior on the way to GA, so treat $2/$12 as provisional.
Can you run both?
Yes, and mature stacks often do. The pattern that works: route by task criticality. The critical path, agentic coding, customer-visible generation, anything where an error is expensive, goes to Opus 4.7. The volume path, ingestion, summarization, media processing, goes to Gemini. A simple router keyed on task type captures most of the savings while protecting the work that justifies flagship spend.
If you'd rather run one model, decide by your dominant workload and accept the tradeoff at the edges.
Which should you pick?
Pick Claude Opus 4.7 if:
- Hard agentic coding is the core of your workload
- Output quality failures are expensive
- You need up to 128K tokens in one response
Pick Gemini 3.1 Pro if:
- Volume economics dominate your decision
- You process video or audio natively
- Preview-stage churn is acceptable for a 55%+ saving