Published per-unit fares look clean because they put unlike jobs on one line. They are also easy to misread. A model can show a launch fare and a list fare, and the gap changes the budget for a bulk image, audio, or text run. The comparison gets messier when a competitor sits between those two numbers. The safe method is to treat launch and list as separate planning cases, then apply the published unit rules, unit minimums, and billing mechanics before submitting a batch. All Y-PAD figures discussed here are provisional beta fares with check dates; confirm the live pricing page before you commit.
Why Launch and List Fares Diverge in Bulk API Budgets
A list fare is the default published rate. A launch fare is a published temporary rate tied to launch seats, not a hidden coupon. Some models have no gap at all during the beta. Y-PAD lists Real-ESRGAN at $0.0010 per image, BiRefNet at $0.0005 per image, Qwen-Image-Edit 2511 at $0.0120 per edited image, and Qwen3-TTS 1.7B at $0.0150 per 1,000 characters, with list and launch the same as of 7 Nov 2026. For those models, launch versus list is a non-issue in the fare table; the real cost drivers are resolution, file count, and per-file minimums.
Other models do split the two fares. Whisper large-v3 is listed at $0.00044 per audio minute on list and $0.0004 on launch. Qwen3-VL 8B is $0.0020 versus $0.0010 per image. GLM-OCR is $0.0010 versus $0.0005 per page. DeepSeek V4.1 Flash is $0.0015 versus $0.0009 per 1,000 weighted tokens. Seedance 2.5 image-to-video and text-to-video are $0.3780 versus $0.1850 per video second at 720p. Pruna P-Video is $0.0210 versus $0.0160 per video second at 720p. WAN 2.6 image-to-video is $0.1050 versus $0.0800 per video second at 720p. A launch fare is not a universal discount; it applies only to the named models when the account holds a launch seat.
The bigger trap is comparing headline fares without normalising the unit. An image, a page, an audio minute, a video second, 1,000 weighted tokens, and 1,000 characters are different denominators. Convert every published fare to a common batch size, such as 1,000 units, and add the unit rules. Whisper large-v3 bills by the second with a 30-second minimum per file, so 1,000 ten-second transcripts cost $0.22 at list and $0.20 at launch, while 1,000 one-minute transcripts cost $0.44 and $0.40. Qwen3-TTS has a 500-character minimum per file, and DeepSeek V4.1 Flash has a 100-token minimum per file. Video fares also carry resolution multipliers: Seedance 2.5 counts 480p at half, 1080p at 2.5×, and 4K at 5×; Pruna P-Video counts 1080p twice; WAN 2.6 counts 1080p at 1.5×. Minimums and multipliers can shrink an apparent launch-to-list saving before the first batch is submitted.
Holds, Seats, and the Real Cost of a Launch Fare
Y-PAD’s billing path matters because the list fare and launch fare do different jobs. At submit, Y-PAD holds the sum of max_units multiplied by the list price. At settlement, succeeded units are charged at the launch price if the account holds a launch seat; otherwise they are charged at list. Unused hold returns, and prices used are stored with the batch. That makes list fare the conservative cash-planning number and launch fare the settlement outcome, not the other way around. Only successful items are charged; failed and cancelled items cost nothing. There is no minimum charge, but the per-file minimums on Whisper, Qwen3-TTS, and DeepSeek still apply.
The launch seat is a limited operational condition, not a permanent price tier. Y-PAD describes 40 launch seats: a $25 pack claims a seat while seats remain, the first $100 of usage is at launch price, and there is one seat per account and per card, assigned at payment. Launch pricing applies to Whisper large-v3, Qwen3-VL 8B, GLM-OCR, DeepSeek V4.1 Flash, Seedance 2.5 image-to-video and text-to-video, Pruna P-Video, and WAN 2.6 image-to-video, at any request size, on succeeded units. The other models have one price during the beta. If seats run out before payment lands, the pack is kept at list prices. First payment cannot exceed $25, and for seven days after first payment the day’s charges and open holds together cannot pass $25. Prepaid balance never expires.
Y-PAD’s own Replicate comparisons add a useful warning. For Real-ESRGAN, BiRefNet, Qwen3-TTS, and Qwen-Image-Edit, the published Replicate figures are above both Y-PAD fares. For Seedance 2.5 at 720p, Replicate is $0.2312 per second, checked 7 Nov 2026, which is below Y-PAD’s $0.3780 list and above its $0.1850 launch. Pruna P-Video has a third-party Replicate listing at $0.02 per second, checked 7 Nov 2026, below Y-PAD’s $0.0210 list and above its $0.0160 launch. So only on the launch-seat fare are those two models cheaper than replicate; at list, they are not. Y-PAD’s tables do not publish Replicate rows for Whisper, DeepSeek V4.1 Flash, GLM-OCR, WAN 2.6, or Qwen3-VL 8B, so do not infer one.
Comparing Published Per-Unit Fares: A Practical Workflow
Build a comparison table before estimating a bulk job. Use columns for model, billing unit, list fare, launch fare, seat eligibility, resolution or duration multiplier, per-file minimum, and check date. Then normalise to a batch size your team understands. At 1,000 units, Real-ESRGAN is $1.00 at both fares and BiRefNet is $0.50 at both. Qwen3-VL 8B is $2.00 at list and $1.00 at launch per 1,000 images, before pro-rata charges above 1 MP up to 4 MP. GLM-OCR is $1.00 versus $0.50 per 1,000 pages, with one image counted as one page and up to 100 pages per file. For DeepSeek V4.1 Flash, remember that the unit is weighted tokens: output tokens plus input tokens at the published 0.25 ratio, with a 100-token minimum per file.
Estimate two quantities: max_units for the hold and expected succeeded units for settlement. Apply multipliers and minimums before multiplying by the fare. A 720p video job may be simple; a 1080p or 4K run is not. A 10-second audio clip is not billed as 10 seconds on Whisper; the 30-second per-file minimum applies. A short TTS line still hits the 500-character minimum. For text jobs, count weighted tokens, not raw words. If the account does not hold a launch seat, use list for both planning and settlement. If it does, use launch only after confirming seat assignment at payment. Because prices used are stored with the batch, a submitted batch keeps its fare even if the table changes later.
Keep the check date beside every figure. Y-PAD’s published prices are provisional beta fares as of 7 Nov 2026, and its Replicate comparisons are mostly checked 2 Nov 2026 or 7 Nov 2026. Before committing a large batch, confirm the live pricing page, the seat status at payment, and the check date on any competitor column. That check is the practical boundary between a published fare and a reliable batch budget.
Busan environmental lawyer now in Montréal advocating river cleanup tech. Jae-Min breaks down micro-plastic filters, Québécois sugar-shack customs, and deep-work playlist science. He practices cello in metro tunnels for natural reverb.