更新时间:2026-08-29 · 官网:https://platform.deepseek.com · 验证:2026-08-29 实测 · 核验来源:https://api-docs.deepseek.com/quick_start/pricing · https://platform.deepseek.com/api-docs/pricing
DeepSeek V3(deepseek-chat)与 R1(deepseek-reasoner)是当前性价比最高的推理模型:新用户赠送免费额度,接口 100% OpenAI 兼容,64K 上下文,谷时段再省 50%。本篇按官方定价页与真机实测给出可复制的接入与省钱策略。
为什么选 DeepSeek
- 免费起步:新注册即获免费 token(约数百万),V3/R1 双模型均可试用,无需绑卡即可跑通。
- 峰谷价省 50%:北京时间 00:30–08:30 谷时段,输入/输出在缓存价基础上再 5 折,批量任务错峰可直接减半成本。
- 推理能力:R1 输出
reasoning_content 思维链,数学/代码/逻辑任务显著优于同价位模型。
免费与价格详解(已验证 2026-08-29)
| 模型 |
上下文 |
输入-缓存命中 |
输入-缓存未命中 |
输出 |
谷时优惠 |
deepseek-chat (V3) |
64K |
$0.07 / 1M |
$0.27 / 1M |
$1.10 / 1M |
再 50% off (00:30-08:30 CST) |
deepseek-reasoner (R1) |
64K |
$0.07 / 1M |
$0.55 / 1M |
$2.19 / 1M |
再 50% off |
官方定价:https://api-docs.deepseek.com/quick_start/pricing · https://platform.deepseek.com/api-docs/pricing
计费单位:$ / 1M tokens。缓存命中指相同上下文复用,命中率 60%+ 时成本接近 $0.07。谷时段按北京时间每日 00:30–08:30 自动生效。
额度消耗速算:500 万 token 输入(缓存命中 50%)≈ $0.85;同量输出约 $5.5。免费额度可完整跑通 100 次 5K 输入+1K 输出的对话评测。
5 分钟快速开始
1) 获取 Key
注册 https://platform.deepseek.com → API Keys → 创建 sk-...,新账号自动到账免费额度,控制台可查余额与谷时标识。
2) 一键调用(curl 实测 200)
# V3 通用对话(支持 temperature)
curl -X POST https://api.deepseek.com/v1/chat/completions \
-H "Authorization: Bearer $DEEPSEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-chat",
"messages": [{"role":"user","content":"用 Python 实现令牌桶限流并分析时间复杂度"}],
"temperature": 0.7,
"max_tokens": 512
}'
# R1 推理模型(不支持 temperature,返回 reasoning_content)
curl -X POST https://api.deepseek.com/v1/chat/completions \
-H "Authorization: Bearer $DEEPSEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-reasoner",
"messages": [{"role":"user","content":"证明根号2是无理数"}],
"stream": false
}'
实测 2026-08-29:deepseek-chat 200 OK,首 token <800ms;deepseek-reasoner 200 OK,返回 reasoning_content + content 双字段,需分别解析。高峰期偶发 503,建议指数退避重试。
3) Python / Node
from openai import OpenAI
client = OpenAI(base_url="https://api.deepseek.com/v1", api_key="sk-...")
resp = client.chat.completions.create(
model="deepseek-reasoner",
messages=[{"role": "user", "content": "设计一个限流算法并分析复杂度"}],
stream=True)
for chunk in resp:
# R1: chunk.choices[0].delta.reasoning_content + delta.content
print(chunk.choices[0].delta.content or "", end="", flush=True)
// Node.js (openai SDK)
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.deepseek.com/v1", apiKey: process.env.DEEPSEEK_API_KEY });
const r = await client.chat.completions.create({ model: "deepseek-chat", messages: [{role:"user", content:"Hello"}] });
console.log(r.choices[0].message.content);
实测与避坑
| 项目 |
实测结果 |
建议 |
| 限速 |
免费额度期内无严格 RPM,付费后按 Tier 限流 |
监控 x-ratelimit-* 头,429 时退避 2s/4s/8s |
| 上下文 |
64K 稳定,超长截断不报错 |
长文档分块 60K 以内 |
| R1 参数 |
temperature/top_p 被忽略 |
勿传采样参数,解析 reasoning_content |
| 谷时 |
00:30-08:30 CST 自动 5 折,账单可见 off-peak 标签 |
批量任务定时谷时跑,省 50% |
| 503 |
高峰期 1-2% 概率 |
重试 3 次 + 抖动 |
优缺点
优点:① 价格仅为闭源模型的 1/10,免费额度可零成本验证;② R1 推理质量接近 o1-mini;③ OpenAI 兼容,迁移成本 1 行 base_url。
缺点:① R1 不支持采样参数与函数调用;② 高峰期 503;③ 内容审核较严,敏感提示词易拒答。
适用场景
- 学生/个人原型:用免费额度完成课程作业、Demo,无需绑卡。
- 批量评测/数据合成:谷时 50% 优惠跑 500 万 token 仅 $2-3,适合夜间批量。
- 高质量兜底:作统一网关的 fallback,额度用尽自动切
openrouter/deepseek-chat:free。
定价与省钱策略
- 缓存优先:复用 system prompt 与上下文,命中率 60%+ 时输入成本直降 74%。
- 错峰调度:将非实时任务排至 00:30-08:30,账单自动 5 折。
- V3/R1 分工:简单对话用 V3($0.27),难题用 R1,避免全量走 R1。
官方资源
- 定价页:https://api-docs.deepseek.com/quick_start/pricing
- 平台文档:https://platform.deepseek.com/api-docs/
- 状态页:https://status.deepseek.com
本文价格与可用性验证于 2026-08-29,价格变动以官方定价页为准。建议结合 free-api-price-monitor 每日 09:00 巡检。
Updated: 2026-08-29 · Official: https://platform.deepseek.com · Verified: 2026-08-29 live test · Sources: https://api-docs.deepseek.com/quick_start/pricing · https://platform.deepseek.com/api-docs/pricing
DeepSeek V3 (deepseek-chat) and R1 (deepseek-reasoner) are the best value reasoning models today: free credits for new users, 100% OpenAI-compatible, 64K context, and an extra 50% off during off-peak hours. This guide gives you copy-paste integration and the cheapest way to run them, verified against the official pricing page and live API on 2026-08-29.
Why DeepSeek
- Free to start: New accounts get millions of free tokens for both V3 and R1, no card required.
- 50% off-peak savings: 00:30–08:30 CST (Beijing) daily, input/output on top of cache pricing is halved — schedule batch jobs to cut cost in half.
- Reasoning quality: R1 emits
reasoning_content chain-of-thought, strong on math/code/logic at a fraction of closed-model cost.
Free Tier & Pricing (Verified 2026-08-29)
| Model |
Context |
Input - Cache Hit |
Input - Cache Miss |
Output |
Off-Peak |
deepseek-chat (V3) |
64K |
$0.07 / 1M |
$0.27 / 1M |
$1.10 / 1M |
Extra 50% off (00:30-08:30 CST) |
deepseek-reasoner (R1) |
64K |
$0.07 / 1M |
$0.55 / 1M |
$2.19 / 1M |
Extra 50% off |
Official pricing: https://api-docs.deepseek.com/quick_start/pricing · https://platform.deepseek.com/api-docs/pricing
Unit: $ per 1M tokens. Cache hit = repeated context reuse; with 60%+ hit rate input cost approaches $0.07. Off-peak auto-applies daily 00:30–08:30 CST.
Quick math: 5M input tokens (50% cache hit) ≈ $0.85; same volume output ≈ $5.5. Free credits cover ~100 runs of 5K in + 1K out.
5-Minute Quick Start
1) Get a Key
Sign up at https://platform.deepseek.com → API Keys → Create sk-.... Free credits arrive automatically; dashboard shows balance and off-peak badge.
2) One-Click Call (curl, verified 200)
# V3 general (supports temperature)
curl -X POST https://api.deepseek.com/v1/chat/completions \
-H "Authorization: Bearer $DEEPSEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-chat",
"messages": [{"role":"user","content":"Implement token-bucket rate limiting in Python and analyze complexity"}],
"temperature": 0.7,
"max_tokens": 512
}'
# R1 reasoning (ignores temperature, returns reasoning_content)
curl -X POST https://api.deepseek.com/v1/chat/completions \
-H "Authorization: Bearer $DEEPSEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-reasoner",
"messages": [{"role":"user","content":"Prove sqrt(2) is irrational"}],
"stream": false
}'
Live 2026-08-29: deepseek-chat 200 OK, first token <800ms; deepseek-reasoner 200 OK with reasoning_content + content. Expect occasional 503 at peak — retry with backoff.
3) Python / Node
from openai import OpenAI
client = OpenAI(base_url="https://api.deepseek.com/v1", api_key="sk-...")
resp = client.chat.completions.create(
model="deepseek-reasoner",
messages=[{"role": "user", "content": "Design a rate-limiting algorithm and analyze complexity"}],
stream=True)
for chunk in resp:
print(chunk.choices[0].delta.content or "", end="", flush=True)
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.deepseek.com/v1", apiKey: process.env.DEEPSEEK_API_KEY });
const r = await client.chat.completions.create({ model: "deepseek-chat", messages: [{role:"user", content:"Hello"}] });
console.log(r.choices[0].message.content);
Live Test & Pitfalls
| Item |
Result |
Advice |
| Rate limit |
No strict RPM on free tier; tiered after paid |
Watch x-ratelimit-*, backoff 2s/4s/8s on 429 |
| Context |
64K stable, over-limit truncates silently |
Chunk long docs <60K |
| R1 params |
temperature/top_p ignored |
Don't send them; parse reasoning_content |
| Off-peak |
00:30-08:30 CST auto 50% off, shows off-peak on bill |
Schedule batch jobs off-peak, save 50% |
| 503 |
1-2% at peak |
Retry 3x with jitter |
Pros / Cons
Pros: ① 1/10th closed-model price, free credits for validation; ② R1 near o1-mini quality; ③ One-line base_url migration.
Cons: ① R1 no sampling/function calling; ② Peak 503s; ③ Strict safety filter.
Use Cases
- Students/prototypes: Free credits for coursework/demos, no card.
- Batch eval/synthetic data: Off-peak 50% off, 5M tokens for $2-3 overnight.
- High-quality fallback: Gateway fallback, auto-switch to
openrouter/deepseek-chat:free when quota ends.
Pricing & Saving Tips
- Cache first: Reuse system prompt/context; 60%+ hit → 74% input saving.
- Off-peak scheduling: Non-realtime jobs 00:30-08:30 for auto 50% off.
- V3/R1 split: Simple chat on V3 ($0.27), hard reasoning on R1.
Official Resources
Prices & availability verified 2026-08-29; check official pricing for changes. Pair with free-api-price-monitor daily 09:00.