Baichuan API pricing is fully out in the open: Baichuan4-Air, the cheapest general model on the lineup, bills combined input+output at just ¥0.98 per 1M tokens, the flagship Baichuan4 reaches ¥100 per 1M tokens, and the M-series (M3/M2) splits input and output pricing in between (source). Baichuan runs two billing styles — most Baichuan4/3/2 models charge a combined input+output token rate, while the M series bills input and output separately. Before comparing prices, figure out which billing style your model uses; never compare a combined price against a split price directly. All prices below are as of the official pricing page on 2026-08-18.
Baichuan API Pricing Comparison: Combined vs Split Billing
Baichuan API pricing can't be read as a single number — combined billing means total tokens × unit price, while split billing requires you to estimate your input/output ratio first. Most Baichuan4/3/2 models (Baichuan4-Air, Baichuan4-Turbo, Baichuan4) charge a single rate on combined input+output tokens, so your bill is simply total token usage times the unit price; the M series (Baichuan-M2/M2-Plus/M3/M3-Plus) prices input and output separately, and you need your input/output mix to compute total cost (source).
The official price page quotes per 1K tokens; this page and article convert to per 1M tokens (×1000): 0.00098 CNY/1K tokens equals ¥0.98/1M tokens, and 1500 tokens bills as 1.5K-token units without rounding. On top of model calls, three add-on fees hit your bill: web_search costs ¥0.03 per call, M3-Plus and M2-Plus chats auto-trigger medical search at ¥0.03 per call, and Knowledge Base file storage bills ¥1.5/GB/day (source).
| Dimension | Combined billing (4/3/2 series) | Split billing (M series) |
|---|---|---|
| Pricing model | Combined input+output tokens × rate | Input and output priced separately |
| Reading bills | Total token usage × unit price | Estimate by input/output ratio |
| Example models | Baichuan4-Air / 4-Turbo / Baichuan4 | Baichuan-M2 / M3 / M3-Plus |
| Best for | Routine Q&A, support, content generation | Long prompts, agents, vertical reasoning |
Full Model Price Table: From ¥0.98 to ¥100 per 1M Tokens
Baichuan's 12 models span more than 100× in price, from Baichuan4-Air at ¥0.98 to the flagship Baichuan4 at ¥100 per 1M tokens — the first step in choosing is always confirming the billing style. Combined-billing general models, lowest to highest (source):
| Model | Billing | Combined ¥/1M tokens | Context | Notes |
|---|---|---|---|---|
| Baichuan4-Air | Combined | 0.98 | 32k | Cheapest general model; officially a new release |
| Baichuan2-Turbo | Combined | 8 | 32k | Legacy integrations |
| Baichuan2-53B | Combined | 10 (00:00-8:00) / 20 (8:00-24:00) | 32k | Time-tiered |
| Baichuan3-Turbo | Combined | 12 | 32k | Prior-gen Turbo |
| Baichuan4-Turbo | Combined | 15 | 32k | Mid Baichuan4 tier |
| Baichuan3-Turbo-128k | Combined | 24 | 128k | Long context; legacy 192k routes here |
| Baichuan4 | Combined | 100 | 32k | Baichuan4 flagship |
The M series bills input/output separately (source):
| Model | Input ¥/1M tokens | Output ¥/1M tokens | Context | Notes |
|---|---|---|---|---|
| Baichuan-M2 | 2 | 20 | 32k | Lowest M-series input price |
| Baichuan-M3-Plus | 5 | 9 | 32k | Auto medical search ¥0.03/call |
| Baichuan-M3 | 10 | 30 | 32k | General reasoning |
| Baichuan-M2-Plus | 10 | 30 | 32k | Same price as M3; auto medical search |
Tools and add-ons (source): Baichuan-Text-Embedding costs ¥0.5 per 1M tokens; Knowledge Base file storage ¥1.5/GB/day with a 5GB cap per user; web_search ¥0.03 per call; Assistants API is free for a limited time.
How to Choose: Baichuan4-Air, M3 or M2
Budget-sensitive routine Q&A goes straight to Baichuan4-Air (¥0.98/1M combined), long-prompt short-generation workloads fit M2, medical verticals need M3-Plus, and for output-heavy traffic compare split prices against Baichuan4-Turbo's combined ¥15 before deciding. Never pick by unit price alone — match your input/output mix to the model: M2's input is only ¥2 but output runs ¥20, ideal for long prompts with short completions; M3-Plus at ¥5 input / ¥9 output pays for its auto medical search (source).
| Your scenario | Recommended model | Billing |
|---|---|---|
| Support, light assistants, budget-sensitive | Baichuan4-Air | Combined ¥0.98/1M tokens |
| Long prompt, short output | Baichuan-M2 | ¥2 input / ¥20 output |
| Healthcare, professional consulting | Baichuan-M3-Plus | ¥5 input / ¥9 output + ¥0.03/search |
| General agents, multi-turn chat | Baichuan-M3 | ¥10 input / ¥30 output |
| Content generation, deep analysis | Baichuan4 | Combined ¥100/1M tokens |
Two quick notes: pass with_search_enhance to disable search on Baichuan4-Air and save ¥0.03 per call when you don't need results; Baichuan2-Turbo-192k is retired and legacy calls route to Baichuan3-Turbo-128k automatically, so existing code needs no changes.
Top-up, Deals and Official Access
Baichuan API is pure pay-as-you-go: the official entry is platform.baichuan-ai.com/prices, and after registering you top up your balance and apply for an API key in the console — there is no monthly subscription. Money goes into your account first, and every call deducts by token usage plus add-ons. The official pricing page also documents registration gift credits — new users registered after 2024-05-22 receive credits valid for 3 months, with the actual amount shown in your console (source). Today the money-savers are Assistants API (free for a limited time) and disabling web_search when you don't need it.
For broader pricing comparisons against other Chinese models, check the DeepSeek, Qwen and Zhipu GLM pages.
Quota, Rate Limits and API Key Setup
Baichuan API quota is driven by your account balance, and the official pricing page publishes no universal rate limits — confirm your console quota and alert rules before a high-concurrency launch. The one hard limit documented is Knowledge Base storage: 5GB per user, billed at ¥1.5/GB/day beyond. Apply for an API key in the Baichuan platform console; text models, embedding, Web Search, Assistants API and Knowledge Base share one key with separate billing rules, and integration follows the official open API docs. This is a pay-as-you-go open interface, not a subscription key.
Is It Worth It: The Bottom Line
For routine Q&A, support or budget-sensitive steady traffic, Baichuan4-Air at ¥0.98 per 1M tokens combined is about as good as it gets; go M3-Plus for medical verticals and Baichuan4 at ¥100 for top-tier quality. The biggest trap in Baichuan pricing is mixing the two billing styles — combined billing reads as total tokens × unit price, split billing needs an input/output estimate. Baichuan reserves the right to change pricing, so factor in add-ons like web_search and medical search when comparing, and always confirm against the official pricing page (source).