What are ads worth in your AI app?

Break-even: 9.9% eligible generations at these settings.

Model prices checked 19 Sep 2026. The default $15 CPM is an illustrative assumption after advertising partner fees and before wavebird’s share, not a measured wavebird rate. External campaign references report advertiser spend on other platforms; they are separate comparisons with a different price basis. The selected starting prices apply to all audience columns; individual prices and fill can then be edited. Fill, qualification, reach and wait times are illustrative. Niche categories have no automatic price premium.

Input includes prompts, history and instructions. Visible output excludes separately entered reasoning. Charge reasoning at the output rate only when that provider bills it that way. Equal token counts compare prices, not model quality or identical text across tokenizers.

Eligibility applies placement, consent and safety rules. Qualification applies the selected event’s validity, visibility or completion checks; CPC adds its valid click rate separately. The illustrative 80% retention gives the second clip 80% of the first clip’s audience before fill and qualification. Two clips is a planning limit, not an industry standard. Longer waits do not automatically increase CPM or click rate.

USD per million tokens. Standard online text inference with uncached input. Token counts apply to every model; use measured billable tokens for your workload. Caching, batch discounts, taxes, regional uplifts and tool charges are excluded unless you add their costs above. Hosted open models use Together AI’s listed API prices, not self-hosting estimates.

This comparison prices text generations only. Image, audio and video workloads need their own cost model. Include allocated storage and processing in other costs, and assess support and acquisition costs for your own margin.

Input is capped at 32,000 tokens and output at 8,000 to stay within the short-context pricing used here. Prices are a checked snapshot, not a live feed. Model listing does not imply a wavebird partnership or assess quality.

Coverage = expected publisher revenue ÷ entered cost for the same generation. Surplus = expected publisher revenue − entered cost. Every generation incurs cost, including those without an ad.

Cost per generation = (input tokens × input price + (visible output + billed reasoning tokens) × output price) ÷ 1,000,000 + other costs per generation.

Expected events = eligible-generation rate × sum of reach across time-eligible positions × fill × qualified share. Each clip starts after its own loading time and the previous clip’s full scheduled duration. CPM starts from rendered opportunities and applies the entered qualification rate. The illustrative vCPM simulation uses an external measurement reference: 50% visibility for 1 continuous second for banners or 2 seconds of video playback. wavebird’s current qualified-view policy uses at least 50% verified visibility for 1 second while the document is visible; the simulation does not change that rule. CPC models valid clicks from rendered opportunities using the entered click rate. CPCV requires full playback plus server validation. Timing alone does not prove an event is valid or payable.

CPM/vCPM revenue = price ÷ 1,000 × expected events × publisher share. CPC revenue = price × valid click rate × expected events × publisher share. CPCV revenue = price per completion × expected completed events × publisher share. Percentages are divided by 100; pricing bases are alternatives, never added together. A stable banner counts at most once. Extra wait adds no events after the last threshold. The selected agreement must support these event definitions.

These timings represent one fixed-duration scenario, not a measured average. For forecasting, evaluate actual wait-time distributions and combine the results; applying event thresholds to an average duration can misstate revenue.

Audience contexts: General chat: Broad everyday queries; diverse advertiser relevance. Film & creation: Editing software, cameras and production tools. Test a higher bid alongside narrower demand. Music: Instruments, audio software and studio equipment. Purchasing intent matters more than the topic alone. Video games: Games, peripherals and hardware. Entertainment browsing and an active purchase can have different value. High-value products: Specialist equipment and expensive products. Test higher bids against a smaller pool of relevant advertisers. Fashion: Clothing and accessories. A specific product search may differ from general style inspiration.

All positions use the same entered price, fill and click rate; real auctions and later-position performance can differ. Clip capacity does not activate a commercial agreement or guarantee independently billable ads. Reasoning and latency are separate inputs; these text-token prices do not price film, music or game asset generation.

Background: Factors affecting ad revenue ↗ · Sequential video ads and duration budgets ↗. These describe general advertising mechanics, not wavebird category rates or partner availability.