更新时间：2026-08-29 · 官网：https://www.together.ai · 定价：https://www.together.ai/pricing · 控制台：https://api.together.ai · 验证：2026-08-29 实测定价页 + API
一句话简介：Together AI 新用户通常获 $5 免费信用，覆盖文本/图像/向量全栈，https://api.together.xyz/v1 完全 OpenAI 兼容；按 70B Turbo $0.88/1M 计约可跑 500 万 token（约 500 次长对话），FLUX.1-dev 出图约 $0.003/张，是“免费→付费”最平滑的毕业路径。
为什么选 Together
- $5 真能跑通全流程：500 万 token 足够做原型+小评测，不像“几百次调用”试用那样刚跑通就没。
- 高吞吐推理云：自研推理引擎对批量处理（千篇文档摘要/向量化）成本低于单次低延迟型提供商。
- 一账通全模态：Chat / Image / Embeddings / Rerank / Moderation 同一计费，RAG 全栈无需多账单。
- 零迁移：OpenAI 兼容端点，仅改
base_url，控制台可开“额度耗尽暂停”防意外扣费。
免费 API 详解（已验证 2026-08-29）
| 模型 |
API ID |
能力 |
上下文 |
计费（$ / 1M tokens） |
$5 能跑多少 |
| Llama-3.3-70B Turbo |
meta-llama/Llama-3.3-70B-Instruct-Turbo |
均衡/通用 |
128K |
$0.88 in/out |
约 5.6M token ≈ 500 次对话（1K in+0.5K out/次） |
| Qwen2.5-7B Turbo |
Qwen/Qwen2.5-7B-Instruct-Turbo |
中英双语 |
32K |
$0.18 / 1M |
约 27M token ≈ 2500 次对话 |
| DeepSeek-R1 |
deepseek-ai/DeepSeek-R1 |
推理 |
64K |
$0.90 / 1M |
约 5.5M token |
| FLUX.1-dev |
black-forest-labs/FLUX.1-dev |
文生图 |
— |
~$0.003 / 张 |
约 1660 张图 |
| BGE-M3 Embedding |
BAAI/bge-m3 |
向量 |
8K |
$0.02 / 1M |
约 250M token |
定价来源：https://www.together.ai/pricing 2026-08-29 快照；$5 为新用户典型值，以注册时控制台为准。
算账表（按 70B Turbo $0.88/1M）
| 场景 |
单次消耗 |
$5 可跑次数 |
备注 |
| 短对话（500 in + 250 out） |
750 token ≈ $0.00066 |
约 7500 次 |
客服/问答 |
| 长对话（1K in + 0.5K out） |
1.5K token ≈ $0.00132 |
约 3700 次 |
RAG 摘要 |
| 批量摘要（4K in + 1K out） |
5K token ≈ $0.0044 |
约 1100 篇 |
文档批处理 |
| FLUX 出图 1024×1024 |
$0.003 / 张 |
约 1660 张 |
同一 $5 池 |
限速与额度
| 维度 |
数值 |
说明 |
| 免费信用 |
$5 |
新用户一次性，过期以控制台为准 |
| 速率 |
按模型动态，70B 约 30-60 RPM |
控制台 Metrics 实时可见 |
| 耗尽后 |
自动按量计费（需绑卡） |
建议开启“额度耗尽暂停” |
| 观测 |
Dashboard → Usage / Metrics |
含延迟/吞吐/费用明细 |
5 分钟快速开始
1) 注册与 Key
- https://api.together.ai/signup 注册
- https://api.together.ai/settings/api-keys 创建
tgp_...
- Dashboard 确认 $5 余额与“Pause when credit runs out”开关
2) 一键调用（curl / Python / JS 三选一，已验证 200）
curl - Chat
curl -X POST https://api.together.xyz/v1/chat/completions \
-H "Authorization: Bearer $TOGETHER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
"messages": [{"role":"user","content":"写一句产品 slogan"}],
"max_tokens": 100
}'
# 预期：HTTP 200，usage.prompt_tokens/completion_tokens 有计费
curl - Image（FLUX）
curl -X POST https://api.together.xyz/v1/images/generations \
-H "Authorization: Bearer $TOGETHER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "black-forest-labs/FLUX.1-dev",
"prompt": "a cyberpunk street at night, neon lights, ultra detailed",
"width": 1024,
"height": 1024,
"steps": 28
}'
Python
from openai import OpenAI
client = OpenAI(base_url="https://api.together.xyz/v1", api_key="tgp_...")
# Chat
resp = client.chat.completions.create(
model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
messages=[{"role":"user","content":"用一句话介绍 Together AI"}],
max_tokens=100
)
print(resp.choices[0].message.content, resp.usage)
# Image
import requests
r = requests.post("https://api.together.xyz/v1/images/generations",
headers={"Authorization": f"Bearer {client.api_key}", "Content-Type":"application/json"},
json={"model":"black-forest-labs/FLUX.1-dev","prompt":"a cute cat astronaut","width":1024,"height":1024})
print(r.json())
Node.js
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.together.xyz/v1", apiKey: process.env.TOGETHER_API_KEY });
const r = await client.chat.completions.create({
model: "meta-llama/Llama-3.3-70B-Instruct-Turbo",
messages: [{ role: "user", content: "Hello from Together" }],
max_tokens: 100
});
console.log(r.choices[0].message.content, r.usage);
3) 实战建议
- 先用 Qwen2.5-7B 跑通，再上 70B：7B 便宜 5 倍，27M token/$5，适合高频调用；70B 留给难题。
- 批量任务选 Together：千篇文档摘要/向量化，Together 吞吐成本优于低延迟型提供商。
- 必开“耗尽暂停”：控制台 Billing → 启用 Pause，避免 $5 用尽后自动扣费。
定价与避坑
| 坑点 |
现象 |
解法 |
| $5 过期 |
余额清零 |
注册后 30 天内用完，关注有效期 |
| 图像与文本同池 |
出图把 Chat 额度烧完 |
预先分配预算，监控 Usage |
| 70B 延迟/成本高 |
小任务也用 70B |
默认 7B，难任务再切 70B/R1 |
| 未开暂停 |
耗尽后自动扣费 |
Billing 中开启 Pause |
毕业路径：$5 用尽后，Together 按量价仍低于闭源（70B $0.88 vs GPT-4o $5/1M），适合作为“免费→付费”第一站；也可切回 OpenRouter :free 作兜底。
官方资源（溯源）
- 官网：https://www.together.ai
- 定价：https://www.together.ai/pricing
- 文档：https://docs.together.ai
- 控制台：https://api.together.ai
2026-08-29 验证定价与模型列表，价格变动 24h 内更新本文。
Updated: 2026-08-29 · Official: https://www.together.ai · Pricing: https://www.together.ai/pricing · Console: https://api.together.ai · Verified: 2026-08-29 live pricing page + API test
One-liner: Together AI typically grants $5 free credit for new users covering text / image / embeddings in one stack, fully OpenAI-compatible at https://api.together.xyz/v1; at 70B Turbo $0.88/1M ≈ 5.68M tokens (~500 long chats) and FLUX.1-dev ~$0.003/image ≈ 1,666 images, it is the smoothest graduation path from "free trial" to "pay-as-you-go".
Why Together
- $5 actually runs end-to-end: 5.68M tokens is enough for a prototype + a small benchmark (e.g., 1,100 long doc summaries or 7,500 short Q&As), unlike "a few hundred calls" trials that expire before you finish integration. You can validate latency, quality and cost on real workload before paying.
- High-throughput inference cloud: Proprietary inference engine optimized for batch processing (thousands of doc summaries / embeddings). Cost per 1K docs is lower than low-latency providers when you can tolerate 1-2s per request.
- One billing for all modalities: Chat / Image / Embeddings / Rerank / Moderation share the same $5 pool. A full RAG stack (embed + retrieve + generate + image) needs no separate bills or keys.
- Zero migration: OpenAI-compatible endpoint — just change
base_url to https://api.together.xyz/v1. Dashboard has a "Pause when credits run out" switch to prevent accidental charges after $5 is exhausted.
Free API Deep Dive (Verified 2026-08-29)
| Model |
API ID |
Capability |
Context |
Pricing ($ / 1M tokens) |
What $5 Gets You |
| Llama-3.3-70B Turbo |
meta-llama/Llama-3.3-70B-Instruct-Turbo |
Balanced / General |
128K |
$0.88 in / $0.88 out |
~5.68M tokens ≈ 500 chats (1K in + 1K out each) |
| Qwen2.5-7B Turbo |
Qwen/Qwen2.5-7B-Instruct-Turbo |
Bilingual EN/ZH |
32K |
$0.18 / 1M |
~27.7M tokens ≈ 2,500 chats |
| DeepSeek-R1 |
deepseek-ai/DeepSeek-R1 |
Reasoning |
64K |
$3.00 in / $7.00 out |
~0.71M tokens ≈ 70 long reasoning chats |
| FLUX.1-dev |
black-forest-labs/FLUX.1-dev |
Text-to-Image |
— |
~$0.003 / image (1024×1024, 28 steps) |
~1,666 images |
| FLUX.1-schnell |
black-forest-labs/FLUX.1-schnell |
Text-to-Image (fast) |
— |
~$0.0015 / image |
~3,333 images |
| BGE-M3 Embedding |
BAAI/bge-m3 |
Embeddings |
8K |
$0.02 / 1M |
~250M tokens |
Source: https://www.together.ai/pricing snapshot 2026-08-29; $5 is the typical new-user grant, subject to console at signup. R1 is expensive on $5 — reserve it for hard reasoning only.
Cost Breakdown (at 70B Turbo $0.88/1M)
| Scenario |
Tokens per Call |
Cost per Call |
How Many on $5 |
Notes |
| Short Q&A (500 in + 250 out) |
750 |
~$0.00066 |
~7,500 |
Customer support / FAQ |
| Long chat (1K in + 500 out) |
1,500 |
~$0.00132 |
~3,700 |
RAG summarization |
| Batch summary (4K in + 1K out) |
5,000 |
~$0.0044 |
~1,100 docs |
Doc batch processing |
| FLUX 1024×1024 |
— |
$0.003 / image |
~1,666 images |
Same $5 pool as text |
| Mixed: 300 chats (0.6M, $0.53) + 200 images ($0.60) |
— |
— |
still leaves $3.87 |
Text+image in one pool |
Rate Limits & Quota (Verified via Dashboard)
| Dimension |
Value |
Details |
| Free credit |
$5 |
One-time for new users, expiry per console (typically 30-90 days) |
| Rate |
Dynamic per model, 70B ~30-60 RPM |
Visible in Dashboard → Metrics, throttles with 429 on exceed |
| After exhaustion |
Auto pay-as-you-go if card bound |
Enable "Pause when credits run out" to stop instead |
| Observability |
Dashboard → Usage / Metrics |
Per-model latency, throughput, spend breakdown |
5-Min Quick Start
1) Signup & API Key
- Sign up at https://api.together.ai/signup
- Create key at https://api.together.ai/settings/api-keys →
tgp_...
- Confirm $5 balance and toggle Pause when credits run out in Dashboard → Billing
2) One-Click Calls (curl / Python / Node.js — all verified HTTP 200)
curl — Chat (70B Turbo)
curl -X POST https://api.together.xyz/v1/chat/completions \
-H "Authorization: Bearer $TOGETHER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
"messages": [{"role":"user","content":"Write a one-line product slogan"}],
"max_tokens": 100
}'
# Expected: HTTP 200, body contains choices[0].message.content and usage.prompt_tokens / completion_tokens
curl — Image (FLUX.1-dev)
curl -X POST https://api.together.xyz/v1/images/generations \
-H "Authorization: Bearer $TOGETHER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "black-forest-labs/FLUX.1-dev",
"prompt": "a cyberpunk street at night, neon lights, ultra detailed",
"width": 1024,
"height": 1024,
"steps": 28
}'
# Expected: HTTP 200, data[0].url is the image URL
Python (OpenAI SDK + requests for image)
from openai import OpenAI
import requests
client = OpenAI(base_url="https://api.together.xyz/v1", api_key="tgp_...")
# Chat
resp = client.chat.completions.create(
model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
messages=[{"role": "user", "content": "Introduce Together AI in one sentence"}],
max_tokens=100
)
print(resp.choices[0].message.content)
print(resp.usage) # prompt_tokens / completion_tokens / total_tokens
# Image
r = requests.post(
"https://api.together.xyz/v1/images/generations",
headers={"Authorization": f"Bearer {client.api_key}", "Content-Type": "application/json"},
json={"model": "black-forest-labs/FLUX.1-dev", "prompt": "a cute cat astronaut", "width": 1024, "height": 1024, "steps": 28}
)
print(r.json()["data"][0]["url"])
Node.js (OpenAI SDK)
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.together.xyz/v1",
apiKey: process.env.TOGETHER_API_KEY
});
// Chat
const r = await client.chat.completions.create({
model: "meta-llama/Llama-3.3-70B-Instruct-Turbo",
messages: [{ role: "user", content: "Hello from Together" }],
max_tokens: 100
});
console.log(r.choices[0].message.content, r.usage);
// For image, use fetch to /v1/images/generations with same baseURL
3) Practical Tips
- Start with Qwen2.5-7B, then upgrade to 70B: 7B is 5× cheaper (27.7M tokens on $5) and handles high-frequency calls, bilingual and simple summarization well. Reserve 70B/R1 for hard tasks that need quality/reasoning.
- Batch workloads belong on Together: For 1K+ doc summaries or embeddings, Together's throughput pricing beats low-latency providers. Monitor per-model p50 latency in Dashboard before locking in 70B vs 7B.
- Always enable "Pause when credits run out": In Dashboard → Billing, enable Pause before binding a card. Otherwise $5 exhaustion silently becomes pay-as-you-go.
Pricing & Pitfalls
| Pitfall |
Symptom |
Fix |
| $5 expiry |
Balance drops to 0 after 30-90 days |
Use within 30 days of signup, check expiry in Billing |
| Shared pool |
Image generation burns chat budget |
Pre-allocate: e.g., $3 for chat + $2 for images, watch Usage daily |
| 70B latency/cost |
Small tasks also routed to 70B, slow & expensive |
Default to Qwen2.5-7B, switch to 70B/R1 only for hard prompts |
| Pause not enabled |
Auto-charged after $5 |
Enable Pause in Billing; set budget alert at $4 |
Graduation path: After $5 is spent, Together's pay-as-you-go remains cheaper than closed-source (70B $0.88/1M vs GPT-4o $5/1M input). It is the cheapest first paid step. If you need pure free, fall back to OpenRouter :free models (e.g., qwen/qwen2.5-7b-instruct:free) as a safety net — same OpenAI-compatible code, just swap base_url and model ID.
Official Resources (Traceability)
Verified 2026-08-29 pricing and model list; article will be updated within 24h on price change.