The lowest rate.
The shortest path.
Model discounts, access and quality across Qianfan MaaS and VOD — with zero data retention on request and a cost model built on real workloads.
Pay a percentage of list. Keep the rest.
Rates are payable percentages of the published international list price. 20% means you pay 20% and save 80% — lower is better.
How to read a rate
The rate is the payable share of the published list price. It assumes standard shared capacity; written SLOs, dedicated clusters, cache-hit targets and data residency are quoted separately. Open-source rates apply to new Qianfan International customers.
100M in × $0.56 = $56.00
20M out × $2.53 = $50.60
list = $106.60
pay 20% = $21.32 · save $85.28
One catalogue. Two product lines.
Qianfan is the OpenAI-compatible MaaS route. VOD is the overseas station with vendor-native APIs and USD billing from Singapore. Where a family sits on both lines, compare the two quotations.
Swipe the table to see every column →
| Model | Rate | List USD / M in · out | Resource basis | Best for |
|---|---|---|---|---|
| ERNIE 5.1 Best rate | 20% | 0.56–0.84 · 2.53–3.10 | Baidu self-deployed | Flagship reasoning and Chinese; 128K context |
| DeepSeek V4 Pro 0813 | 30% | 1.32 · 3.96 (cache 0.132) | Official upstream | Deep reasoning and code; 1M context |
| DeepSeek V4 Flash 0731 | 30% | 0.44 · 1.32 | Official upstream | High-throughput assistants and extraction |
| GLM 5.3 | 70% | On request | Official upstream | Agentic and multi-step work; strongest GLM quality |
| GLM 5.3 Flash | 40% | On request | Token factory | High-volume agents at the lowest GLM rate |
| GLM 5.2 | 30% | 1.40 · 4.40 | Official upstream | Bilingual workloads; 500K context |
| GLM 5.2 Fast | 50% | 2.25 · 7.88 | Official upstream | Latency-optimised sibling of GLM 5.2 |
| Kimi K3 | 60% | On request | Official upstream | Long-context documents, repositories, transcripts |
| Qwen 3.7 Plus | 50% | On request | Official upstream | Bilingual, cost-sensitive workloads |
| Gemini · all series | 60% | On request | Official key | Multimodal frontier |
| GPT chat / response | 55% | On request | Official key (Microsoft) | Frontier text and response APIs |
| Claude | 60% | On request | Official routed — standard | Frontier reasoning; 70% premium, 90% official key |
Open-source rates (DeepSeek, GLM, Kimi, Qwen) apply to new Qianfan International customers. “On request” means the model is priced at launch or in the formal quotation; no price is estimated.
| Model | Rate | Resource basis | Best for |
|---|---|---|---|
| Qianfan-VL V1 | 50% | Baidu official | Visual understanding at the lowest vision rate |
| Qianfan-VL V1.1 | 100% | Baidu official | New vision model; list rate applies |
| Qianfan-VL V1.1 Highspeed | 100% | Baidu official | Latency-optimised variant of V1.1 |
| ERNIE 4.5 Turbo VL | On request | Baidu official | Visual understanding at published list price |
| GPT image 2 / 2.5 | 55% | Microsoft official | Image generation and editing; capacity per project |
Vision models are billed per 1M input and output tokens; image models per image. Availability is confirmed at quotation.
| Model | Rate | Resource basis | Best for |
|---|---|---|---|
| Qianfan-video V7 | 51% | Official routed · SD 2.0 Mini | Full asset library with real-person detection |
| Qianfan-video V6 | 61% | SD 2.0 Fast | Faster generation for iteration |
| Qianfan-video V8 | 75% | H3-Max | Top quality tier |
| Qianfan-Sota V1 | 48% | New generation | Newest generation engine |
| HappyHorse Lowest | 35% | Official key | Lowest unit cost in video |
| PixVerse (before V6) | 45% | Official key | Cost-sensitive social clips |
| Vidu · all series | 50% | Official key | High-quality generation at mid cost |
| Kling · all series | 55% | Official key | Quality tier for campaign shots |
| Wan 3.0 | 50% | Official key | General-purpose generation |
| Hailuo H02 / H23 | 45% | Official key | Cost-effective clips |
| Seedance 2.0 | 75% | Full version | Draft, original-URL and egtd workflows |
| MiniMax TTS | 40% | Per 1M characters | Voice for dubbing and narration |
How a video quote is built
Generation time = billable seconds × list price per second × rate. Concurrency is a separate capacity premium set by peak concurrent tasks. Delivery — BOS storage, transcoding, moderation and CDN — is quoted by VOD.
$10,000 list generation
× 51% = $5,100 payable
concurrency & delivery quoted separately
| Model | SKU | VOD product ID | Billing |
|---|---|---|---|
| Gemini | — | S17235954403197 | Per token |
| Claude | Official key | S17235954403206 | Per token |
| Claude | Official routed — high | S17235954403207 | Per token |
| Claude | Official routed — low | S17235954403211 | Per token |
| GPT chat | — | S17235954403194 | Per token |
| GPT image 2 / 2.5 | — | S17235954403183 | Per image |
| Kimi K3 | — | S17235954403190 | Per token |
| GLM 5.1 / 5.2 | — | S17235954403180 | Per token |
| GLM 5.3 / 5.3 Flash | — | S17235954403181 | Per token |
| DeepSeek V4 Pro | — | S17235954403189 | Per token |
| DeepSeek V4 Flash | — | S17235954403177 | Per token |
| DeepSeek V3.2 | — | S17235954403199 | Per token |
One VOD key is bound to a region but not a model: the same Singapore key calls Claude, GPT, Gemini, Kimi, GLM and DeepSeek, and the charge lands on the matching product ID. Product IDs are the order key. Storage, transcoding, moderation and CDN are separate VOD line items.
Discounts that ship with the pipeline.
Media-processing tools are billed through VOD and discounted under partner policy — stack them with generation and delivery on one quote.
List prices for generated media are set per model, resolution and duration; the exact unit for your project is confirmed in the quotation. USD settlement, Singapore billing.
What the rate is worth.
Price a text workload against published list. Change the model, volume or cache-hit rate to see the monthly and annual difference.
What every quotation includes.
The rate is only useful if the model passes your evaluation and the platform holds the load. These are the controls that travel with it.
USD settlement
International accounts bill in USD. VOD settles from Singapore; Qianfan keys are enabled by the commercial team.
Zero data retention
ZDR on request: whitelist review, signed addendum, locked configuration and an acceptance test. ZDR does not change the rate.
99.9% SLA
Multi-region failover and 24/7 enterprise support on qualifying services; written SLOs quoted separately.
Official channels only
No pooled or non-official capacity, and no model distillation without prior written approval.
Capacity
Billion-TPM class across self-deployed, upstream and token-factory supply. Reserved TPM and concurrency quoted separately.
Flexible billing
Pay-as-you-go, committed volume or subscription, with credit top-ups for creative tools.
Rates follow the 26 October 2026 schedule and are payable percentages of published list price. They are indicative, subject to capacity checks, regional availability and approval, and may change with one week's notice. USD list; RMB settlement at USD × 7.2. Final availability, billing unit, discounts, taxes, SLA and legal terms are confirmed in an official quotation. This page is not an offer or a guarantee of service.
Three numbers are enough to start.
Monthly volume, peak concurrency and the guarantee you need. We confirm the model, the resource basis and a firm rate — with ZDR if you need it.