One transparent rate card
Language-model prices are USD per 1 million tokens and include XinoAPI gateway service. Featured prices use 90% of the matching Baidu Qianfan International list rate. Input and cache-hit input are shown separately where available.
| Model | Input / cache hit (per 1M) | Output (per 1M) | Context |
|---|---|---|---|
DeepSeek V4-Flash Featured −10% deepseek-v4-flash · 1M context, 384K output, dual reasoning modes | Input $0.1400$0.1260 Cache hit $0.0280$0.0252 | $0.2800$0.2520 | 1M |
DeepSeek V4-Pro Featured −10% deepseek-v4-pro · 1.6T MoE, SWE-bench 80.6%, approaches Claude Opus | Input $1.690$1.521 Cache hit $0.140$0.126 | $3.380$3.042 | 1M |
Qwen Turbo qwen-turbo · fast and cheap, good for high volume | $0.051$0.061 | $0.204$0.245 | 1M |
Qwen Plus New qwen-plus · Qwen3.7 generation, balanced quality, 1M context | $0.111$0.133 | $0.667$0.800 | 1M |
Qwen Max Flagship qwen-max · Qwen3.7 Max, top coding benchmarks | $0.333$0.400 | $1.333$1.600 | 256K |
GLM-5.3 Latest flagship glm-5.3 · Zhipu flagship, 1M context, 128K maximum output | Input $1.192 Cache hit $0.298 | $4.172 | 1M |
GLM-5.2 Featured −10% glm-5.2 · previous generation, 1M context, selected-value route | Input $1.400$1.260 Cache hit $0.260$0.234 | $4.400$3.960 | 1M |
GLM-5.1 glm-5.1 · previous generation flagship, 203K context | $0.97$1.164 | $3.04$3.648 | 203K |
Kimi K3 New kimi-k3 · 2.8T MoE, frontier coding, 1M context | $3.00$3.600 | $15.00$18.000 | 1M |
Kimi K2.6 kimi-k2.6 · cost-effective tier, 256K context | $0.903$1.083 | $3.75$4.500 | 256K |
MiniMax M3 New MiniMax-M3 · 428B MoE, MSA architecture, 1M context, multimodal | $0.292$0.350 | $1.167$1.400 | 1M |
Dola Seed 2.1 Turbo New dola-seed-2-1-turbo · ByteDance Seed / Doubao family, multimodal understanding | Input $0.500 Cache hit $0.100 | $2.500 | 256K |
On mobile, swipe the table horizontally to view all input, output, and context rates.
Strikethrough on featured rows shows the comparable Baidu Qianfan International list rate, published September 7, 2026 and verified September 17, 2026. Other routes remain at their currently published XinoAPI rates pending individual review.
Video pricing in its native unit
Video generation is billed by successful output seconds or generation tokens, depending on the provider. Failed generations are not billed where the upstream provider reports no billable output.
| Model | Meter | Published rate | Output |
|---|---|---|---|
MiniMax H3 Native audio minimax-h3 · text, image, video, and audio references | Successful output duration | 768P $0.080/sec 2K $0.130/sec | 4–15 sec · 24fps |
Wan 3.0 Video New wan3.0-video · all-in-one text, image, reference, editing, and extension | Successful output duration | 480P $0.056/sec 720P $0.112/sec 1080P $0.224/sec | 2–30 sec · 30fps |
Dreamina Seedance 2.5 New dreamina-seedance-2-5-260628 · long-form multimodal video generation | Generation tokens | Without video input $10.70/M With video input $6.40/M | 480P / 720P · 4–30 sec |
On mobile, swipe the table horizontally to compare meters, rates, and output limits.
Sources verified September 17, 2026: MiniMax H3 pricing, Wan 3.0 international pricing, and BytePlus ModelArk pricing. Wan USD rates are fixed launch rates rounded from the Singapore international list using the September 16 reference exchange rate. MiniMax H3 bills input video at the selected output-resolution rate; audio is free and the first five reference images are free. Seedance charges actual generation tokens reported by the provider.
What real workloads cost
Common usage shapes at XinoAPI prices, estimated per month.
Chat Assistant · 500 users/day
Code Assistant · 10 devs
Document Summary · bulk
Reasoning Workflow
Free to start, no credit card
Eligible new accounts receive $2.00 in one-time trial credits. That covers roughly 7.9M DeepSeek V4-Flash output tokens at the current published rate, or any equivalent published usage.
Scale without hidden cuts
Published featured-model savings apply automatically. Enterprise pricing is reviewed against actual monthly usage, model mix, capacity, and support needs.
For enterprise volumes (>$5K/month), invoicing, or dedicated routing, contact sales.
Pricing is not the whole decision
XinoAPI is designed for users outside mainland China and routes requests to third-party model providers with different data policies.
| Control | Current policy | Why it matters |
|---|---|---|
| Mainland China access | Not permitted for registration, purchase, dashboard access, or API use. | Maintains a clear cross-border service boundary for Chinese LLM inference export. |
| Prompt/response storage | No plaintext content retention by default; billing uses metadata such as model, tokens, status, and timestamps. | Reduces data exposure for production agent and application workloads. |
| Provider terms | Users must comply with each upstream provider's terms, data policy, and regional restrictions. | XinoAPI is a gateway, not the developer or operator of upstream models. |
| Sensitive data | Use the Privacy SDK for local PII and secret redaction before sending prompts. | Provider-side policies vary, especially for models operated in mainland China. |
See the Compliance Center and Security Whitepaper for the full policy.
Pricing questions, answered
pip install xinoapi-privacy.deepseek-v4-flash for general tasks (chat, RAG, code completion) and deepseek-v4-pro for flagship reasoning, planning, and code review. Use explicit V4 model IDs in new projects; avoid relying on legacy DeepSeek aliases.