Sora 2 vs Sora 2 Pro
什么时候选哪个
sora-2-pro 每输出秒贵 3 倍($0.3 对 $0.1),换来的是一档分辨率:720p 或 1024p,对比只有 720p。其余完全一致——4–12 秒、24 fps、16:9 和 9:16、原生音频,以及单张参考图的图生视频。除非交付格式需要更高的画幅,否则选 sora-2。
Benchmark 成绩
Sora 2:供应商没有公布 benchmark 成绩。
Sora 2 Pro:公布了 6 项成绩,但每一项测过的其他模型都太少,没法比较。
供应商公布: ByteDance
价格
| Sora 2 | Sora 2 Pro | Δ | |
|---|---|---|---|
| 每秒输出(带音频) | $0.1 | $0.3 | 0.33× |
价格取自构建时的实时目录,最新价格见各模型页面。
两个模型所处的位置:全部 7 个按同一单位计费的视频生成模型的每秒输出价格分布(对数刻度)
能力
| Sora 2 | Sora 2 Pro | |
|---|---|---|
| 原生音频 | 是 | 是 |
规格
| Sora 2 | Sora 2 Pro | |
|---|---|---|
| 输入模态 | 文本 图像 | 文本 图像 |
| 输出模态 | 视频 | 视频 |
| 发布日期 | 2025-09 | 2025-09 |
| 分辨率 | 720p | 720p / 1024p (固定选项) |
| 视频时长 | 4-12 s (区间) | 4-12 s (区间) |
| 帧率 | 24 fps | 24 fps |
| 宽高比 |
|
|
| 输入方式 |
|
|
| 格式 | .mp4 | .mp4 |
规格照录自各供应商的文档;供应商没有公布的项目,对应的行直接省略,不做推断。 完整来源: Sora 2 · Sora 2 Pro
同一条提示词,两个模型,经网关实测
模型返回 720×1280 时长 4s 延迟 68 s
模型返回 1280×720 时长 4s 延迟 130 s
同一条提示词,每个模型只请求一次,不重试,也不挑结果,展示的都是各模型返回的第一个结果。尺寸和时长都没有指定,各模型用的是自己的默认值,因为硬凑一组所有模型都接受的参数,哪个模型都发挥不好。这里的文件为了网页展示重新编码过,所以请看构图和对提示词的还原程度,不要看压缩画质。
改一行代码就能在两个模型之间切换
下方每个标签页里都有两个模型 ID,高亮的那两行是唯一要改的地方。端点不变,API key 不变,请求结构也不变。
import time
import requests
BASE = "https://synthorai.io/v1"
HEADERS = {"Authorization": "Bearer sk-syn-..."}
# 1) create the video generation job (POST /v1/videos)
task = requests.post(
f"{BASE}/videos",
headers=HEADERS,
json={
"model": "sora-2",
# "model": "sora-2-pro", # 取消注释此行,注释上一行
"prompt": "a watercolor lighthouse at dawn, waves rolling in, camera pulling back",
"resolution": "720p",
"duration": 5,
},
).json()
# 2) poll the job until it reaches a terminal state
while task["status"] in ("queued", "in_progress"):
time.sleep(5)
task = requests.get(f"{BASE}/videos/{task['id']}", headers=HEADERS).json()
# 3) completed → signed video URL (valid ~24h - download and store it promptly)
if task["status"] == "completed":
print(task["data"][0]["url"])
else:
print(task["error"])const HEADERS = {
Authorization: "Bearer sk-syn-...",
"Content-Type": "application/json",
};
interface VideoTask {
id: string;
status: "queued" | "in_progress" | "completed" | "failed" | "cancelled";
data?: Array<{ url: string }>;
error?: { code: string; message: string };
}
// 1) create the video generation job (POST /v1/videos)
const created = await fetch("https://synthorai.io/v1/videos", {
method: "POST",
headers: HEADERS,
body: JSON.stringify({
model: "sora-2",
// model: "sora-2-pro", // 取消注释此行,注释上一行
prompt: "a watercolor lighthouse at dawn, waves rolling in, camera pulling back",
resolution: "720p",
duration: 5,
}),
});
let task = (await created.json()) as VideoTask;
// 2) poll the job until it reaches a terminal state
while (task.status === "queued" || task.status === "in_progress") {
await new Promise((r) => setTimeout(r, 5000));
const res = await fetch(`https://synthorai.io/v1/videos/${task.id}`, { headers: HEADERS });
task = (await res.json()) as VideoTask;
}
// 3) completed → signed video URL (valid ~24h — download and store it promptly)
if (task.status === "completed") console.log(task.data?.[0]?.url);
else console.error(task.error);# Create the job; "Prefer: wait" holds the request up to 60s and returns the
# completed task when generation finishes in time (else the queued task).
curl https://synthorai.io/v1/videos \
-H "Authorization: Bearer sk-syn-..." \
-H "Content-Type: application/json" \
-H "Prefer: wait=60" \
-d '{
"model": "sora-2",
# "model": "sora-2-pro", # 取消注释此行,注释上一行
"prompt": "a watercolor lighthouse at dawn, waves rolling in, camera pulling back",
"resolution": "720p",
"duration": 5
}'
# Still queued / in_progress? Poll until completed → data[0].url (valid ~24h).
curl https://synthorai.io/v1/videos/vid_9f2e8c1a4b7d \
-H "Authorization: Bearer sk-syn-..."package main
import (
"bytes"
"encoding/json"
"fmt"
"net/http"
"time"
)
const base = "https://synthorai.io/v1"
type videoTask struct {
ID string `json:"id"`
Status string `json:"status"`
Data []struct {
URL string `json:"url"`
} `json:"data"`
}
func call(method, url string, body []byte) (t videoTask) {
req, _ := http.NewRequest(method, url, bytes.NewReader(body))
req.Header.Set("Authorization", "Bearer sk-syn-...")
req.Header.Set("Content-Type", "application/json")
resp, _ := http.DefaultClient.Do(req)
defer resp.Body.Close()
json.NewDecoder(resp.Body).Decode(&t)
return t
}
func main() {
// 1) create the video generation job (POST /v1/videos)
payload, _ := json.Marshal(map[string]any{
"model": "sora-2",
// "model": "sora-2-pro", // 取消注释此行,注释上一行
"prompt": "a watercolor lighthouse at dawn, waves rolling in, camera pulling back",
"resolution": "720p",
"duration": 5,
})
task := call("POST", base+"/videos", payload)
// 2) poll the job until it reaches a terminal state
for task.Status == "queued" || task.Status == "in_progress" {
time.Sleep(5 * time.Second)
task = call("GET", base+"/videos/"+task.ID, nil)
}
// 3) completed → signed video URL (valid ~24h — download and store promptly)
if task.Status == "completed" {
fmt.Println(task.Data[0].URL)
}
}import java.net.URI;
import java.net.http.HttpClient;
import java.net.http.HttpRequest;
import java.net.http.HttpResponse;
// JDK built-in HttpClient — no extra dependency needed.
HttpClient http = HttpClient.newHttpClient();
// 1) create the job; "Prefer: wait=60" holds the request up to 60s and
// returns the completed task when generation finishes in time
HttpRequest create = HttpRequest.newBuilder(URI.create("https://synthorai.io/v1/videos"))
.header("Authorization", "Bearer sk-syn-...")
.header("Content-Type", "application/json")
.header("Prefer", "wait=60")
.POST(HttpRequest.BodyPublishers.ofString("""
{"model": "sora-2",
// {"model": "sora-2-pro", // 取消注释此行,注释上一行
"prompt": "a watercolor lighthouse at dawn, waves rolling in, camera pulling back",
"resolution": "720p", "duration": 5}"""))
.build();
String task = http.send(create, HttpResponse.BodyHandlers.ofString()).body();
System.out.println(task); // {"id":"vid_…","status":…} — completed → data[0].url
// 2) still queued / in_progress? Poll GET https://synthorai.io/v1/videos/{id} until the
// status turns completed, then download data[0].url (valid ~24h).常见问题
Sora 2 和 Sora 2 Pro 哪个更便宜?
按「每秒输出(带音频)」算,Sora 2 更便宜($0.1 对 $0.3,相差 3.0×)。其他计费项的结论可能相反,完整价格见上方表格,实际成本取决于你的用量构成。
不用分别集成两次,就能对 Sora 2 和 Sora 2 Pro 做 A/B 测试吗?
可以。两个模型走同一个 OpenAI 兼容端点,用同一个 API key,切换时只要改一行里的模型名,所以可以给两个模型各分一部分流量,直接对比账单。
它们支持哪些视频时长和分辨率?
上方的规格表格列出了各模型公布的分辨率和视频时长,并标明取值是固定选项还是连续区间。这两种看起来差不多,花的钱却不一样,因为按秒计费时,价格要乘以时长。