together.ai
THE AI NATIVE CLOUD

Build what's next on the AI Native Cloud.

Train, fine-tune, and serve open-source models on the fastest inference stack — at 10× lower cost than the hyperscalers. From a hosted endpoint to a dedicated GPU cluster.

Available in 14 regions SOC 2 · HIPAA H100 · H200 · B200
AI natives build on Together AI
SALESFORCE NEXUSFLOW DUOLINGO ARCEE.AI WASHINGTON POST ZOOM PIKA
THE PLATFORM

The Together AI Platform.

Full-stack infrastructure for AI workloads — from serverless inference to dedicated training clusters. Drop-in compatible with the OpenAI client. No vendor lock-in.

10×
Lower inference cost vs. hyperscaler endpoints
200+
Open-source models · serverless and dedicated
99.9%
Uptime SLA · enterprise tier
Inference Fine-tuning GPU Clusters Evaluations
SERVERLESS INFERENCE

One API. Two hundred models.

Llama, Mistral, Qwen, DeepSeek, Stable Diffusion, FLUX, and every major open-source release — behind a single endpoint that's drop-in compatible with the OpenAI client. Pay only for the tokens you use.

Read the docs View model catalog
PYTHON inference.py — together / serverless
1# Drop-in compatible with the OpenAI client
2from together import Together
3
4client = Together()
5
6response = client.chat.completions.create(
7  model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
8  messages=[{"role": "user", "content": "Hello"}],
9  stream=True,
10)
11
12for chunk in response:
13  print(chunk.choices[0].delta.content or "", end="")
RESEARCH

Grounded in cutting-edge research.

The team behind FlashAttention, RedPajama, and the Together Inference Engine. We publish what we ship, and we ship what scales.

INFERENCE

The Together Inference Engine

An end-to-end stack that delivers 3-4× higher throughput than vLLM on the same hardware, with sub-100ms first-token latency on Llama 3.3 70B.

v3.2GA · Dec '25
DATA

RedPajama-v2

30T tokens of curated pretraining data with full provenance. Used to train Mistral, Mixtral, and a generation of open-source frontier models.

DATASET30 trillion tokens
ATTENTION

FlashAttention-3

The H100-optimized attention kernel that ships in every major training and inference framework. Open source, MIT-licensed.

PAPERNeurIPS '24
PRICING

Transparent per-token pricing.

Pay only for what you generate. No volume tiers, no marketing seats. Switch to a dedicated endpoint when you outgrow serverless.

TEXT VISION IMAGE AUDIO VIDEO
STANDARD WHOLESALE
MODEL CONTEXT INPUT OUTPUT STATUS
Llama 3.3 70B Instruct Turbo
meta-llama / 70B
128K $0.88/ 1M tok $0.88/ 1M tok ON-DEMAND
DeepSeek V3
deepseek-ai / 671B MoE
64K $1.25/ 1M tok $1.25/ 1M tok ON-DEMAND
Mixtral 8×22B Instruct
mistralai / 141B MoE
64K $1.20/ 1M tok $1.20/ 1M tok ON-DEMAND
Qwen 2.5 72B Instruct
alibaba / 72B
32K $0.90/ 1M tok $0.90/ 1M tok ON-DEMAND
Llama 3.1 8B Instruct Turbo
meta-llama / 8B
128K $0.18/ 1M tok $0.18/ 1M tok ON-DEMAND
Get started now View full catalog All prices in USD. Wholesale tier available for 50M+ tokens/mo.
WHAT'S NEW

From the Together blog.

INFERENCE

The Together Inference Engine v3.2 — sub-100ms first-token latency on Llama 3.3 70B.

How we cut tail latency on the most-served open model by 38% with speculative decoding, prefix caching, and a redesigned scheduler.

CUSTOMERS

How Pika ships frontier video models on 4,096 H200s.

A look inside Pika's training pipeline — 32-node clusters, gradient checkpointing, and the schedule that lets the research team prioritize without breaking SRE budgets.

DESIGN SYSTEM · V.ALPHA

The system in the open.

Tokens and primitives behind the Together surface — colors grouped by role, the display-sans + uppercase-mono ladder, the lightly-rounded radius scale, and the canonical component set. Source: DESIGN.md. Dark hero, white middle, gradient ribbon as the single piece of chrome.

01 · Color

Black CTA. Dark-light alternating bands. Three-stop gradient as decoration.

Brand & Gradient
primary / ink
#000000
accent-orange
#fc4c02
accent-magenta
#ef2cc1
accent-periwinkle
#bdbbff
accent-mint
#c8f6f9
brand gradient
3-stop
Surface
canvas
#ffffff
canvas-dark
#010120
surface-dark-soft
#313641
surface-dark-bord
#26263a
hairline
#ebebeb
hairline-strong
#d4d4d4
Text
ink
#000000
body
#959494
on-dark
#ffffff
on-dark-muted
#b6b6c4
02 · Typography

Display sans for narrative. Uppercase mono for every label.

display-xxl
64 / 500 / -1.92
Build what's next on the AI Native Cloud.
display-xl
40 / 500 / -0.8
The Together AI Platform.
display-lg
28 / 500 / -0.42
Grounded in cutting-edge research.
display-md
22 / 500 / -0.22
Card titles, research-card headings.
body-lg
18 / 400 / -0.18
Lead paragraphs sit slightly larger than body.
body-md
16 / 400 / -0.16
Default body paragraph copy.
body-md-strong
16 / 500 / -0.16
Bolded inline body emphasis.
mono-caps-button
Mono / 16 / 500 / 0.08
GET STARTED NOW
mono-caps-eyebrow
Mono / 11 / 500 / 0.55
SECTION EYEBROW
caption
14 / 400 / 0
Caption for fine print.
mono-caption
Mono / 10 / 400 / 0.05
BY THE INFRA TEAM · 2026-04-12
03 · Radius

4px is the canonical radius. Pills only for the chat orb.

none
0
xs
3.25
sm
4px
md
8px
full
9999
04 · Components

Black CTA pill. Mint and white reserved for hero. Hairline cards.

Buttons

Mono uppercase labels. Black is the canonical CTA. Mint and white live in hero contexts.

Inputs

4px radius, hairline-strong border. Inverts to ink border on focus.

Cards

Hairline 4px on white. Surface-dark-bord 4px on dark. No shadows.
ARTICLE-CARD
Light surface, hairline border.
Used for testimonials and blog cards.
RESEARCH-CARD
Dark surface on the research band.
Mono eyebrow + display headline + body.

Tags & badges

All caps mono. Light hairline default. Gradient tag is the brand chrome at chip scale.
DEFAULT ON-DEMAND NEW FRONTIER
STANDARD WHOLESALE
together.ai
VIEW DESIGN.MD ↗