Chinese LLMs and AI agents, all pricing in one table
Pricing for Chinese LLMs and AI agents built for individuals, collected from each vendor's official pricing page. This page is a periodically updated snapshot and may lag behind live prices.
246246 models · 27 productsUpdated
Pick models and compare
Compare input, output, cache-hit prices, subscription plans and free tiers at a glance. Identical values are merged across columns.
¥4.5Cache hit ¥0.15Peak hours: Input ¥9 · Cache hit ¥0.3 · Output ¥27
¥6Cache hit ¥1.2
Output¥/1M
¥13.5
¥30
Subscription
Subscription plans
—
Coding Plan Lite · 1 month¥40/mo
Coding Plan Lite · 3 months¥120/mo
Coding Plan Pro · 1 month¥200/mo
Coding Plan Pro · 3 months¥600/mo
Agent Plan Small¥40/mo
Agent Plan Medium¥200/mo
Agent Plan Large¥500/mo
Agent Plan Max¥1,000/mo
Free tier
Free on the chat site / app
Ark new users get about 1M tokens; first activation of Managed Agents includes 30 hours of Agent runtime + 500 web_search calls
Notes
Note
Use deepseek-flash as the model name. Legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still callable, but those models are retired; requests are served by DeepSeek-V4.1-Flash and billed at Flash rates. Context length 1M, max output 384K.; The API uses peak and off-peak pricing: peak hours (Mon–Fri 09:00–12:00 and 14:00–18:00 Beijing time, excluding Chinese public holidays) cost 2×; off-peak is half the peak rate. All other times (including weekends and public holidays) are off-peak.
No fixed public monthly fee for the app
Chinese LLM API & subscription pricing comparison
Sort by API input / output price or lowest subscription. Click a column header to toggle ascending or descending. Data updated .
Disclaimer: this is a curated snapshot and may differ from live pricing. Always confirm on each vendor's official site.
Data summary
Tracking 12 products and 246 models. Rows are split by model; the main input price is the cache-miss rate, with cache-hit, tiered and promotional prices shown in the cell.
Among CNY-priced models, the lowest input price is ¥0.125 per 1M tokens (Qwen · qwen3.7-text-embedding-flash).
The lowest output price is ¥0 per 1M tokens (MiMo · mimo-v2.5-asr, Qwen · qwen-mt-uni).
The lowest individual monthly plan is ¥15/month (iFlytek Spark · App Membership).
USD-priced products (LongCat) are shown with $ at face value, with no currency conversion and excluded from CNY price rankings.
¥1Cache hit ¥0.02Peak hours: Input ¥2 · Cache hit ¥0.04 · Output ¥8
¥4
—
Free on the chat site / app
Use deepseek-flash as the model name. Legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still callable, but those models are retired; requests are served by DeepSeek-V4.1-Flash and billed at Flash rates. Context length 1M, max output 384K.; The API uses peak and off-peak pricing: peak hours (Mon–Fri 09:00–12:00 and 14:00–18:00 Beijing time, excluding Chinese public holidays) cost 2×; off-peak is half the peak rate. All other times (including weekends and public holidays) are off-peak.Show note
Promotional credits / new-account quota; TTS models (mimo-v2.5-tts, mimo-v2.5-tts-voiceclone, mimo-v2.5-tts-voicedesign) are free for a limited time and do not consume plan Credits; cache writes are free for a limited time
Supports Batch calls at 50% of the standard rate (input 1.5, cache hit 0.0125, output 3); For pay-as-you-go billing in China, use a standard API Key from the open platform; usage is deducted from your account balance based on actual Token usage and is not interchangeable with Token Plan quotas. Cache hits are billed at the cache hit price; cache writes are free for a limited time. Batch inference is priced at 50% off the standard real-time inference price. Batch inference is not supported for: mimo-v2.6-pro-ultraspeed; mimo-v2-pro, mimo-v2-omni, and mimo-v2-tts were discontinued on 2026-06-30; mimo-v2.5-pro and mimo-v2.5 were officially discontinued at 10:00 Beijing time on 2026-10-21. We recommend switching to the new model versions as soon as possible. Off-peak hours (0:00-8:00 Beijing time, i.e. UTC 16:00-24:00) have a 0.8x consumption multiplier. The ASR series is billed by input audio duration (mimo-v2.5-asr: ¥0.5/hour in China, $0.074/hour overseas). Web search is billed separately per call (¥16/1000 calls in China, $5/1000 calls overseas) and is not included in Token pricing.Show note
Flagship, 1M context; cache write TTL 5 minutes ¥20 / 1 hour ¥40; supports batch inference, fees at 60% of standard price; Kimi API automatically enables context cache billing for repeated request prefixes; cache TTL defaults to 5 minutes. For the K3 series, cache writes are billed separately by TTL (5 minutes / 1 hour), and cache-hit input is billed only at the cache hit price. kimi-k3.1-pro and kimi-k3 support batch inference (Batch API), with all fees at 60% of standard prices. The K2 series models will stop being available to newly registered users starting October 31, 2026, and all K2 series models are planned to be discontinued by the end of December.Show note
Coding model, 262K context; no longer available to newly registered users starting October 31, 2026, planned to be discontinued by the end of DecemberShow note
High-speed version of K2.7 Code, 262K context; no longer available to newly registered users starting October 31, 2026, planned to be discontinued by the end of DecemberShow note
Supports vision and text input, 262K context; no longer available to newly registered users starting October 31, 2026, planned to be discontinued by the end of DecemberShow note
Priority service is billed at 1.5× the standard rate; set service_tier=priority when calling; An API price increase has been announced; refer to live official pricingShow note
GLM-5.3 has been released (API coming soon; currently only available via GLM Coding Plan); for GLM-5.2 domestic pricing, refer to live open-platform rates; flagship cache hits are free for a limited time; GLM-5.3-Flash is 50% off for two weeks; Batch API is 50% of standard pricing; cache storage is free for most models for a limited time; GLM-TTS is billed at ¥2 per 10K charactersShow note
¥12Cache hit ¥1Peak hours: Input ¥9 · Cache hit ¥0.3 · Output ¥27Off-peak hours: Input ¥4.5 · Cache hit ¥0.15 · Output ¥13.5
¥24
Peak/Off-peak hours are limited-time pricing for DeepSeek-V4-Pro-0813 (Cache hit limited-time pricing effective dates: September 9–October 30)Show note
Flagship reasoning model; cache hit price adjusted from 0.35 CNY per 1M tokens to 0.7 CNY per 1M tokens from 2026-10-14 00:30; users who have already made pay-as-you-go calls continue to enjoy 0.35 CNY per 1M tokens; Non-token billing: Text-to-speech 580/280/90 CNY per 1M characters, voice cloning and voice design 10 CNY per voice, speech recognition 0.45–5.5 CNY per hour, StepAudio 3 series billed by session minute/billed character/audio hour, text-to-image 0.1/0.02 CNY per image; value-added capabilities: web search 0.04 CNY per call, image search by text 0.12 CNY per call, file storage 0.5 CNY/GB/day. StepAudio 3 series free for a limited time until 2026-11-17 11:59:59 Beijing time; billed at new prices from 2026-11-17 12:00:00.Show note
¥0.65 per session minute (¥39.00 per session hour)
—
Free for a limited time until 2026-11-17 11:59:59 Beijing time; billed by session duration from 2026-11-17 12:00:00, with the smallest billing unit being 1 second and any duration under 1 second counted as 1 secondShow note
Pricing basis: rows are split by model. The main input price is the cache-miss rate; cache-hit, tiered and promotional prices appear in smaller text. LongCat is priced in USD and shown with $ directly, with no currency conversion. Pricing changes frequently. Always confirm on the vendor's live pricing page before use.
FAQ
Which Chinese LLM has the lowest API input price?
Among CNY-priced models, the lowest input price is ¥0.125 per 1M tokens (Qwen · qwen3.7-text-embedding-flash). Cache-hit and tiered prices are shown in the corresponding table cells.
Which Chinese LLM has the lowest output price?
Based on the data tracked here, the lowest output price is ¥0 per 1M tokens (MiMo · mimo-v2.5-asr, Qwen · qwen-mt-uni).
Which Chinese LLM has the cheapest individual subscription?
The lowest CNY individual monthly plan tracked here is ¥15/month (iFlytek Spark · App Membership).
How do USD-priced models compare with CNY prices?
This site does not convert USD prices. USD-priced products (LongCat) are shown at face value with $ in a separate column; do not compare the raw numbers directly.
How often is the data updated?
Data updated 2026-11-11. A scheduled job fetches vendor pricing pages every 3 days. Prices change often; confirm on the official site before purchasing.