cli and products
Together AI
Together AI — US-based inference + training platform for open-source models. Serverless per-token, dedicated endpoints, fine-tuning, GPU clusters. Hosts Kimi K3, DeepSeek V4, GLM-5.x, MiniMax M3, Qwen, Llama.
## what it does
Overview
Together AI is a US-based inference and training platform (San Francisco) hosting open-source models — DeepSeek V4 Pro 0813 / Flash 0731, Kimi K3, GLM-5.3/5.3-Flash/5.2, MiniMax M3, Qwen3.8 2.4T A95B, GPT OSS 120B, Llama 3.3 70B, plus image / audio / video / embedding / moderation families. The Python SDK and the OpenAI-compatible REST API are both first-class.
Why Together AI?
- US data residency. Use DeepSeek / Qwen / GLM weights without routing through China.
- OpenAI-compatible REST at
https://api.together.ai/v1— drop-in for OpenAI clients. - Serverless for prototyping; dedicated endpoints for prod latency; GPU clusters with on-demand B200s for training and large batch jobs.
→ See DeepSeek Privacy Guide
Selected serverless prices (fetched 2026-09-10)
Source: together.ai/pricing. Per 1M tokens, USD.
| Model | Input | Cached input | Output |
|---|---|---|---|
| MiniMax M3 | $0.30 | $0.06 | $1.20 |
| Kimi K3 | $3.00 | $0.30 | $15.00 |
| DeepSeek V4 Flash 0731 | $0.14 | $0.03 | $0.28 |
| DeepSeek V4 Pro 0813 | $1.32 | $0.13 | $3.96 |
| Qwen3.8-2.4T-A95B | $2.00 | $0.25 | $6.00 |
| Qwen3.7-Plus | $0.32 | — | $1.28 |
| Qwen3.7-Max | $1.25 | $0.13 | $3.75 |
| Qwen3.8 Flash | $0.15 | — | $0.47 |
| Qwen3.6-Plus | $0.50 | — | $3.00 |
| GLM-5.3 | $1.40 | $0.26 | $4.40 |
| GLM-5.3-Flash | $0.15 | $0.03 | $0.50 |
| GLM-5.2 | $1.40 | $0.26 | $4.40 |
| GPT OSS 120B | $0.15 | — | $0.60 |
| Muse Glimmer 30B | $0.35 | $0.04 | $1.50 |
| Inkling | $1.00 | $0.17 | $4.05 |
| Llama 3.3 70B | $1.04 | — | $1.04 |
| Gemma 4 31B | $0.39 | — | $0.97 |
| Cogito v2.1 671B | $1.25 | — | $1.25 |
| LFM2.5-8B-A1B (Liquid) | $0.03 | — | $0.12 |
| Ternary Bonsai 27B | $0.00 | — | $0.00 |
Display prices are lowest resolution/duration settings. Multimodal prices (image, audio, video) follow the same page under their respective tabs.
Quick start
from together import Together
client = Together()
completion = client.chat.completions.create(
model="moonshotai/Kimi-K3",
messages=[{"role": "user", "content": "What are the top 3 things to do in New York?"}],
)
print(completion.choices[0].message.content)cURL works the same way:
curl -X POST "https://api.together.ai/v1/chat/completions" \
-H "Authorization: Bearer $TOGETHER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"moonshotai/Kimi-K3","messages":[{"role":"user","content":"Top 3 things to do in NY?"}]}'Surfaces
| Surface | Use it for |
|---|---|
| Serverless Inference | Pay-per-token, no setup. Cheapest path. |
| Provisioned Throughput | Reserved capacity at predictable cost. |
| Dedicated Inference | Single-tenant GPUs for production isolation. |
| GPU Clusters | On-demand B200s with attached storage for training or large batch. |
| Fine-Tuning | LoRA / full-param SFT + DPO on launch models. |
| Sandbox / Managed Storage / Batch | Code execution, persistent storage, async batch jobs. |
Compliance
- SOC 2 Type II certified
- Data residency: US (San Francisco)
- No data training: Your data is not used for model training
- Data retention: Configurable, default 30 days
Related
- DeepSeek — Why access via Together instead of direct
- Fireworks AI — Alternative US inference platform
- DeepSeek Privacy Guide — Complete privacy walkthrough
- Best models by lab
## embed this badge
cache ✓ in cache.directory
