LLM API prices & free tiers
The same question can cost 50× apart
APIs bill per token: flagship output runs over ¥100 per million tokens, while Chinese value tiers cost a few yuan. Most individuals don't need the flagship — check free tiers and subscriptions before reaching for the API.
Last updated
Price table (per million tokens)
| Model | Vendor | Input | Output | Free tier | Subscription | Checked |
|---|---|---|---|---|---|---|
| GPT-6.1 Sol ↗ | OpenAI | $1.25 / ¥8.9 | $10 / ¥71 | 无免费 API 额度（ChatGPT 免费版可网页用） | ChatGPT Plus $20/月（Plus 内含 GPT-6.1 限额） | 2026-11-07 |
| GPT-6 Astra ↗ | OpenAI | $2.5 / ¥17.8 | $20 / ¥142 | 无免费 API 额度 | ChatGPT Pro $200/月（Pro 档全量） | 2026-11-07 |
| Claude Opus 5.5 ↗ | Anthropic | $5 / ¥35.5 | $25 / ¥177.5 | 无免费 API 额度（claude.ai 免费档限次） | Claude Pro $20/月 / Max $100–200/月 | 2026-11-07 |
| Claude Sonnet 5.5 ↗ | Anthropic | $3 / ¥21.3 | $15 / ¥106.5 | 无免费 API 额度 | Claude Pro $20/月（Sonnet 为主力档） | 2026-11-07 |
| Claude Haiku 5.5 ↗ | Anthropic | $1 / ¥7.1 | $5 / ¥35.5 | 无免费 API 额度 | Claude Pro $20/月 | 2026-11-07 |
| Gemini 4 Carbon ↗ | $2.5 / ¥17.8 | $15 / ¥106.5 | AI Studio 免费档限速可用（限额浮动） | Google AI Pro ¥139.99/月（大陆区以 Play 结算为准） | 2026-11-07 | |
| Gemini 3.8 Flash ↗ | $0.3 / ¥2.1 | $2.5 / ¥17.8 | AI Studio 免费档限速可用 | Google AI Pro ¥139.99/月 | 2026-11-07 | |
| DeepSeek V4.1 ↗ | DeepSeek（深度求索） | $0.28 / ¥2 | $0.42 / ¥3 | 无免费额度（价格即开即用，无最低消费） | 无订阅制，纯 API 按量 | 2026-11-07 |
| DeepSeek R1（推理档） ↗ | DeepSeek（深度求索） | $0.42 / ¥3 | $2 / ¥14 | 无免费额度 | 无订阅制，纯 API 按量 | 2026-11-07 |
| Kimi K3.1 Pro ↗ | Moonshot AI（月之暗面） | $0.6 / ¥4 | $2.5 / ¥18 | 新用户赠送额度（以开放平台公告为准） | Kimi 会员 ¥49/月起（网页/App 端） | 2026-11-07 |
| GLM-5.3 ↗ | Z.ai（智谱） | $0.28 / ¥2 | $0.84 / ¥6 | glm-5.3-flash 档免费（限速） | GLM Coding Plan ¥20/月起 | 2026-11-07 |
| Qwen3.8 Max ↗ | Alibaba（阿里云百炼） | $2.25 / ¥16 | $9 / ¥64 | 各模型新用户免费 token 额度（百万级，90 天有效） | 无统一订阅，按量计费 | 2026-11-07 |
| Doubao Seed 1.8（豆包） ↗ | ByteDance（火山方舟） | $0.35 / ¥2.5 | $1.1 / ¥8 | 每款模型 50 万 token 免费额度 | 无统一订阅，按量计费 | 2026-11-07 |
| ERNIE 5.0（文心） ↗ | Baidu（百度千帆） | $1.1 / ¥8 | $4.2 / ¥30 | 千帆新用户赠送调用次数（以控制台为准） | 文心一言 App 会员 ¥59.9/月（App 端） | 2026-11-07 |
| MiniMax M3.1 ↗ | MiniMax（稀宇科技） | $0.42 / ¥3 | $1.7 / ¥12 | 新用户赠送额度（以平台公告为准） | 无统一订阅，按量计费 | 2026-11-07 |
| Hunyuan Turbo S（混元） ↗ | Tencent（腾讯云） | $0.7 / ¥5 | $2.8 / ¥20 | 各模型限时免费额度（以控制台为准） | 无统一订阅，按量计费 | 2026-11-07 |
- GPT-6.1 Sol ↗2026-11-07
OpenAI
- Input
- $1.25 / ¥8.9
- Output
- $10 / ¥71
Free tier: 无免费 API 额度（ChatGPT 免费版可网页用）
Subscription: ChatGPT Plus $20/月（Plus 内含 GPT-6.1 限额）
- GPT-6 Astra ↗2026-11-07
OpenAI
- Input
- $2.5 / ¥17.8
- Output
- $20 / ¥142
Free tier: 无免费 API 额度
Subscription: ChatGPT Pro $200/月（Pro 档全量）
- Claude Opus 5.5 ↗2026-11-07
Anthropic
- Input
- $5 / ¥35.5
- Output
- $25 / ¥177.5
Free tier: 无免费 API 额度（claude.ai 免费档限次）
Subscription: Claude Pro $20/月 / Max $100–200/月
- Claude Sonnet 5.5 ↗2026-11-07
Anthropic
- Input
- $3 / ¥21.3
- Output
- $15 / ¥106.5
Free tier: 无免费 API 额度
Subscription: Claude Pro $20/月（Sonnet 为主力档）
- Claude Haiku 5.5 ↗2026-11-07
Anthropic
- Input
- $1 / ¥7.1
- Output
- $5 / ¥35.5
Free tier: 无免费 API 额度
Subscription: Claude Pro $20/月
- Gemini 4 Carbon ↗2026-11-07
Google
- Input
- $2.5 / ¥17.8
- Output
- $15 / ¥106.5
Free tier: AI Studio 免费档限速可用（限额浮动）
Subscription: Google AI Pro ¥139.99/月（大陆区以 Play 结算为准）
- Gemini 3.8 Flash ↗2026-11-07
Google
- Input
- $0.3 / ¥2.1
- Output
- $2.5 / ¥17.8
Free tier: AI Studio 免费档限速可用
Subscription: Google AI Pro ¥139.99/月
- DeepSeek V4.1 ↗2026-11-07
DeepSeek（深度求索）
- Input
- $0.28 / ¥2
- Output
- $0.42 / ¥3
Free tier: 无免费额度（价格即开即用，无最低消费）
Subscription: 无订阅制，纯 API 按量
- DeepSeek R1（推理档） ↗2026-11-07
DeepSeek（深度求索）
- Input
- $0.42 / ¥3
- Output
- $2 / ¥14
Free tier: 无免费额度
Subscription: 无订阅制，纯 API 按量
- Kimi K3.1 Pro ↗2026-11-07
Moonshot AI（月之暗面）
- Input
- $0.6 / ¥4
- Output
- $2.5 / ¥18
Free tier: 新用户赠送额度（以开放平台公告为准）
Subscription: Kimi 会员 ¥49/月起（网页/App 端）
- GLM-5.3 ↗2026-11-07
Z.ai（智谱）
- Input
- $0.28 / ¥2
- Output
- $0.84 / ¥6
Free tier: glm-5.3-flash 档免费（限速）
Subscription: GLM Coding Plan ¥20/月起
- Qwen3.8 Max ↗2026-11-07
Alibaba（阿里云百炼）
- Input
- $2.25 / ¥16
- Output
- $9 / ¥64
Free tier: 各模型新用户免费 token 额度（百万级，90 天有效）
Subscription: 无统一订阅，按量计费
- Doubao Seed 1.8（豆包） ↗2026-11-07
ByteDance（火山方舟）
- Input
- $0.35 / ¥2.5
- Output
- $1.1 / ¥8
Free tier: 每款模型 50 万 token 免费额度
Subscription: 无统一订阅，按量计费
- ERNIE 5.0（文心） ↗2026-11-07
Baidu（百度千帆）
- Input
- $1.1 / ¥8
- Output
- $4.2 / ¥30
Free tier: 千帆新用户赠送调用次数（以控制台为准）
Subscription: 文心一言 App 会员 ¥59.9/月（App 端）
- MiniMax M3.1 ↗2026-11-07
MiniMax（稀宇科技）
- Input
- $0.42 / ¥3
- Output
- $1.7 / ¥12
Free tier: 新用户赠送额度（以平台公告为准）
Subscription: 无统一订阅，按量计费
- Hunyuan Turbo S（混元） ↗2026-11-07
Tencent（腾讯云）
- Input
- $0.7 / ¥5
- Output
- $2.8 / ¥20
Free tier: 各模型限时免费额度（以控制台为准）
Subscription: 无统一订阅，按量计费
Exchange rate: 1 USD ≈ 7.10 CNY (2026-11-07). For overseas models the CNY figure is converted (≈); for Chinese vendors the CNY figure is the official list price and USD is back-calculated (≈).
Prices change frequently — always confirm on the linked official pricing page before buying. This table is manually verified and refreshed every two weeks.
A few terms first
Subscription or API?
Personal daily use: subscribe first. ChatGPT Plus / Claude Pro / Kimi membership run about ¥140/month and include far more usage than the same budget in API credits — right for chat, writing and research.
Developer integration / batch jobs: use the API. Programmatic calls, embedding into your own product and batch runs require it; there the question shifts from "strongest" to "most tokens per yuan", where Chinese value tiers are usually enough.
Heavy agent users: do the math. One agent task can burn hundreds of thousands of tokens — a flagship run can outcost a month of subscription. Validate the workflow on a cheaper tier before upgrading.
FAQ
Which LLM APIs have free tiers?
Chinese vendors broadly do: GLM has a free model tier (glm-5.3-flash, rate-limited), Doubao grants 500k tokens per model, Alibaba Bailian gives new users million-token credits, and MiniMax and Baidu Qianfan both offer sign-up grants. Overseas, Google AI Studio's free tier is the most generous (rate-limited access to flagship models); OpenAI and Anthropic official APIs have none. Free-tier policies change often — confirm on the official page.
Why is output several times the input price?
Output tokens are generated one at a time (autoregressive), which costs far more compute than input tokens, which can be processed in parallel. Output is typically 3–10× input. For coding and long-form writing, compare output prices first.
How are the CNY figures computed?
Overseas models are converted at the 2026-11-07 rate (1 USD ≈ 7.10 CNY); "≈" marks approximations — your payment channel settles the real rate. For Chinese vendors the yuan figure is the official list price and the dollar figure is back-calculated. This table is re-verified every two weeks; public price changes are updated immediately.
Why isn't every model listed?
The table covers what most people actually buy: each vendor's flagship, its value workhorse and a free-tier representative. Long-tail variants (monthly-refresh SKUs, regional editions) change too fast to maintain honestly — a stale listing misleads more than an omission. Full lists live on the vendor pages.
About the data
This column cites rather than reproduces: we quote only key scores and findings, each labeled with its source, version date and a link to the original — never a full-table copy of anyone's leaderboard.
Automated sources (LMArena / SWE-bench / LiveBench) sync weekly; the Artificial Analysis intelligence index and the pricing table are manually verified, with check dates noted in the changelog.
Benchmarks are not your experience. Each board measures something different and scores drift between releases — test against your own task before committing.
各厂商官方定价页（人工整理）Version 2026-11-07License Factual price information compiled manually from official vendor pricing pages; prices change frequently — always verify on the linked official page before purchasing.各厂商官方定价页（人工整理）
Written by the Flowerclaw editorial team · Reviewed by the Flowerclaw editorial team