China Model API Relay Market Monthly · 2026 年 10 月
One issue per month. Figures come from daily automated measurement of 2132 confirmed relay sites; judgments and analysis are written by Sinan Lab, based only on these measurements, and are not a recommendation of any channel.
Our reading of the month
本期分析撰写中。
一、规模与结构
| Panel family | Sites | Open reg. | Close | Unknown | With public quotes | Median ratio |
|---|---|---|---|---|---|---|
| one-api | 918 | 702 | 103 | 113 | 541 | 15% |
| sub2api | 823 | 667 | 89 | 67 | 0 | — |
| multimodal-aggregator | 236 | 77 | 20 | 139 | 0 | — |
| one-api/new-api | 95 | 67 | 10 | 18 | 55 | 27% |
| openai-compatible | 43 | 23 | 2 | 18 | 0 | — |
| openai-compatible? | 7 | 4 | 0 | 3 | 0 | — |
Panel family is determined by public API fingerprint; Sub2API-family plan prices sit behind login, so there is no public quote this issue.
| Login method | one-api family | Sub2API family |
|---|---|---|
| 641 | 807 | |
| GitHub | 240 | 122 |
| LINUX DO | 134 | 46 |
| 0 | 96 | |
| OIDC | 38 | 0 |
| Discord | 26 | 0 |
| Telegram | 17 | 0 |
| 11 | 0 |
| TLD | Sites |
|---|---|
| .com | 643 |
| .ai | 226 |
| .top | 184 |
| .cc | 158 |
| .xyz | 108 |
| .cn | 106 |
| .net | 69 |
| .vip | 43 |
| Source | Sites this period |
|---|---|
| content_fp | 132 |
| ct_log+mass_probe | 29 |
| Directory: batch | 25 |
| adjacency | 20 |
| github | 18 |
| feeds | 12 |
| submission | 2 |
| telegram | 1 |
二、价格
| Model | Vendor | Tier | Official $/M | Relay median $/M | Ratio | Sites |
|---|---|---|---|---|---|---|
| Grok 4.7 | xai | Mid | 6.00 | 0.225 | 4% | 246 |
| GPT-6.1 Sol | openai | Mid | 10.00 | 0.600 | 6% | 430 |
| GPT-6.1 Terra | openai | Mid | 5.00 | 0.300 | 6% | 219 |
| GPT-6 Astra | openai | Flagship | 50.00 | 3.00 | 6% | 476 |
| GPT-6 Sol | openai | Mid | 10.00 | 0.600 | 6% | 430 |
| Claude Opus 5.5 | anthropic | Flagship | 20.00 | 2.25 | 11% | 384 |
| Claude Sonnet 5.5 | anthropic | Mid | 10.00 | 1.33 | 13% | 324 |
| Gemini 3.8 Flash | Mid | 4.12 | 0.562 | 14% | 252 | |
| Gemini 4 Argon | Mid | 10.00 | 1.50 | 15% | 199 | |
| Claude Fable 5.1 | anthropic | Flagship | 50.00 | 8.32 | 17% | 330 |
| Claude Haiku 5.5 | anthropic | Fast | 0.500 | 0.110 | 22% | 209 |
| GPT-6 Luna | openai | Fast | 0.500 | 0.135 | 27% | 227 |
| MiniMax M3.1 Flash | minimax | Fast | 0.300 | 0.090 | 30% | 63 |
| GLM-5.2 | zhipu | Mid | 7.00 | 2.14 | 31% | 246 |
| Qwen 3.8 Flash | alibaba | Fast | 0.470 | 0.182 | 39% | 121 |
| GLM-5.3 Flashx | zhipu | Fast | 1.25 | 0.525 | 42% | 64 |
| Qwen 3.8 Max | alibaba | Mid | 6.00 | 2.97 | 49% | 184 |
| Kimi K3 | moonshot | Mid | 15.00 | 7.50 | 50% | 269 |
| GLM-5.3 | zhipu | Mid | 3.60 | 2.10 | 58% | 243 |
| MiniMax M3 | minimax | Fast | 1.20 | 0.756 | 63% | 167 |
| DeepSeek V4.1 Flash | deepseek | Fast | 0.500 | 0.360 | 72% | 268 |
| Qwen 3.7 Plus | alibaba | Fast | 1.28 | 0.925 | 72% | 115 |
| Qwen 3.7 Max | alibaba | Mid | 4.42 | 3.24 | 73% | 142 |
| DeepSeek V4 Pro | deepseek | Fast | 1.90 | 1.42 | 75% | 314 |
| Qwen 3.7 Flash | alibaba | Fast | 0.130 | 0.105 | 81% | 24 |
| GLM-5.3 Flash | zhipu | Fast | 0.132 | 0.210 | 159% | 223 |
| DeepSeek V4 Flash | deepseek | Fast | 0.182 | 0.360 | 198% | 266 |
Discount ratio = relay market median effective price ÷ official reference price; domestic / overseas by the vendor's country of registration.
| Tier | Quotes | Below cost floor | Below bulk discount | Within range | Near public | Above public | Far above |
|---|---|---|---|---|---|---|---|
| Flagship | 1183 | 64% | 20% | 4% | 7% | 3% | 2% |
| Mid | 3108 | 53% | 17% | 10% | 9% | 6% | 4% |
| Fast | 1968 | 20% | 22% | 13% | 14% | 18% | 14% |
Band definitions are in section 3 of the definitions page: <15% below cost floor · 15–40% below common bulk discount · 40–80% within range · 80–125% near public price · 125–200% above · >200% well above.
| Top-up rate (CNY per $1 nominal) | Sites | Share |
|---|---|---|
| 1 | 371 | 62% |
| 7.3 | 111 | 19% |
| 7 | 19 | 3% |
| 6.8 | 15 | 3% |
| 0.1 | 9 | 2% |
| 0.2 | 5 | 1% |
| 0.25 | 5 | 1% |
| 0.5 | 4 | 1% |
Each site's most common top-up rate; 594 sites with public quotes. 1.0 = ¥1 buys $1 of quota (effective price about 15% of official); 6.8–7.3 ≈ real exchange rate 1:1.
Largest moves in market median this period
| Model | Start → end $/M | Change | Sites · % of official |
|---|---|---|---|
| Qwen 3.7 Flash Fast | $5.20 → $0.105 | -98% | 24 sites · 81% |
| MiniMax M3.1 Flash Fast | $0.374 → $0.090 | -76% | 60 sites · 15% |
| Grok 4.7 Mid | $0.891 → $0.225 | -75% | 243 sites · 4% |
| Claude Haiku 5.5 Fast | $0.301 → $0.107 | -64% | 204 sites · 21% |
| GPT-6 Astra Flagship | $7.42 → $3.00 | -60% | 470 sites · 6% |
| GPT-6.1 Terra Mid | $0.224 → $0.337 | +50% | 199 sites · 7% |
| Gemini 4 Argon Mid | $1.46 → $1.50 | +2% | 193 sites · 15% |
| Qwen 3.7 Plus Fast | $1.19 → $0.924 | -22% | 113 sites · 72% |
| GPT-6 Luna Fast | $0.178 → $0.135 | -24% | 220 sites · 27% |
| MiniMax M3 Fast | $1.01 → $0.755 | -25% | 164 sites · 63% |
3. Reachability and probes
| Eligible sites | 7-day 100% | 99%–99.9% | Below 99% | Our outage rounds excluded |
|---|---|---|---|---|
| 499 | 145 | 154 | 200 | — |
| Lowest reachability | 7-day reach | Probes | Handshake p50 |
|---|---|---|---|
| comput.chat | 45.2% | 168 | 1518ms |
| kg-api.cloud | 50.9% | 167 | 3301ms |
| mxdapi.com | 55.5% | 146 | 3038ms |
| feiyuanapi.com | 61.9% | 168 | 1886ms |
| waterapi.vip | 73.0% | 148 | 2124ms |
| Probe | Site×model pairs | Consistent | Differs | Capability pairs | Below median |
|---|---|---|---|---|---|
| This period | 47 | 0 | 0 | 13 | 0 |
| Audit · hold reason | Count |
|---|---|
| Leader gap | 6 |
| Lone outliers | 9 |
| Unit hints | 152 |
| unit_suspect | 295 |
| Released this period | 0 |
4. Image and video
| Video family | Sites | Comparable quotes | Official ref $/s | Family median $/s | Median ÷ ref | Below discount / above public |
|---|---|---|---|---|---|---|
| Wan (Alibaba) | 40 | 187 | $0.045 | $0.105 | 233% | 48 / 81 |
| Seedance (ByteDance) | 72 | 173 | $0.100 | $0.123 | 123% | 26 / 74 |
| Kling (Kuaishou) | 29 | 145 | $0.084 | $0.113 | 134% | 16 / 75 |
| Veo（Google） | 58 | 139 | $0.100 | $0.075 | 75% | 93 / 10 |
| Hailuo / H3（MiniMax） | 63 | 101 | $0.080 | $0.066 | 82% | 22 / 29 |
| Vidu (Shengshu) | 29 | 83 | $0.042 | $0.070 | 167% | 1 / 46 |
| Grok Imagine Video (xAI) | 83 | 82 | $0.050 | $0.019 | 39% | 53 / 10 |
| Image family | Sites | Comparable quotes | Official ref $/image | Family median $/image | Median ÷ ref | Below discount / above public |
|---|---|---|---|---|---|---|
| Nano Banana / Gemini Image（Google） | 173 | 563 | $0.067 | $0.023 | 34% | 355 / 47 |
| Seedream (ByteDance) | 61 | 217 | $0.033 | $0.025 | 76% | 85 / 37 |
| Grok Imagine Image (xAI) | 108 | 190 | $0.020 | $0.013 | 64% | 97 / 40 |
| Qwen-Image (Alibaba) | 45 | 141 | $0.027 | $0.029 | 109% | 40 / 55 |
| Wan Image (Alibaba) | 32 | 37 | $0.075 | $0.016 | 21% | 19 / 9 |
| Kling Image (Kuaishou) | 4 | 5 | $0.028 | $0.0058 | 21% | 3 / 2 |
| FLUX（BFL） | 42 | 4 | $0.100 | $0.022 | 22% | 0 / 0 |
The reference price uses the lowest official tier (resolution / duration); relay quotes are converted to the same tier; families without a public official price get no ratio.
5. Compute cost layer
| GPU | Vast.ai on-demand median $/h | RunPod secure $/h |
|---|---|---|
| RTX 4090 | 0.44 | 0.89 |
| RTX 4090 D | — | — |
| RTX 5090 | 0.59 | 1.19 |
| RTX 5090 D | — | — |
| RTX A6000 | 0.46 | 0.59 |
| RTX PRO 6000 | — | — |
| L40S | 0.80 | 1.09 |
| V100 32GB | — | — |
| Open model | Official $/M | Relay median $/M | Ratio | Whether the official price is below the H100 single-card cost line (about $0.222/M) |
|---|---|---|---|---|
| Qwen 3.7 Flash | 0.130 | 0.105 | 81% | Yes |
| GLM-5.3 Flash | 0.132 | 0.210 | 159% | Yes |
| DeepSeek V4 Flash | 0.182 | 0.360 | 198% | Yes |
| MiniMax M3.1 Flash | 0.300 | 0.090 | 30% | No |
| Qwen 3.8 Flash | 0.470 | 0.182 | 39% | No |
| DeepSeek V4.1 Flash | 0.500 | 0.360 | 72% | No |
| MiniMax M3 | 1.20 | 0.756 | 63% | No |
| GLM-5.3 Flashx | 1.25 | 0.525 | 42% | No |
| Qwen 3.7 Plus | 1.28 | 0.925 | 72% | No |
| DeepSeek V4 Pro | 1.90 | 1.42 | 75% | No |
| GLM-5.3 | 3.60 | 2.10 | 58% | No |
| Qwen 3.7 Max | 4.42 | 3.24 | 73% | No |
| Qwen 3.8 Max | 6.00 | 2.97 | 49% | No |
| GLM-5.2 | 7.00 | 2.14 | 31% | No |
| Kimi K3 | 15.00 | 7.50 | 50% | No |
The cost line is a rough estimate under the assumptions in section 13 of the methodology doc (H100 $2.40/h, 3000 tokens/s), for order of magnitude only.
Definitions and corrections
All metric definitions are in Definitions; raw data:2026-10.json、price-index.json、data_v2.json. Figures are recalculated daily during the month, finalized at month end and marked "final"; the analysis text is updated together with the final version.
Cite this page
Data may be freely cited and republished with attribution and a link. We take no money from measured channels, sell no rankings and add no referral parameters, so you can cite us as an independent third-party measurement.
司南实验室，《中国模型 API 中转市场月报 · 2026 年 10 月》，2026-11-01，https://compute.sinanlab.com/report/2026-10（访问日期 ____）Sinan Lab, "中国模型 API 中转市场月报 · 2026 年 10 月", 2026-11-01, https://compute.sinanlab.com/report/2026-10 (accessed ____)Media, research institutions and financial institutions needing raw data or custom definitions: write to hello@sinanlab.com, see Press & researchers。
Subscribe to the Sinan weekly
This week's rankings, the price index, mainstream price changes, newly indexed and unreachable sites. Measurements only, no recommendations; unsubscribe any time.
Clear direction, grounded judgement.