inference

The world's most powerful
AI models cheaper, faster.

The world's most powerful
AI models cheaper, faster.

Run Inference

One API. Low latency. Data-resident. Drop-in compatible with the OpenAI and Anthropic SDKs you already use — switch by changing one line.

One API. Low latency. Data-resident. Drop-in compatible with the OpenAI and Anthropic SDKs you already use — switch by changing one line.

Screenshot of the ai& Playground answering a question about hosting AI in Japan, generated in 70 tokens at 38 milliseconds using moonshotai/kimi-k2.6.
Screenshot of the ai& Playground answering a question about hosting AI in Japan, generated in 70 tokens at 38 milliseconds using moonshotai/kimi-k2.6.
Bar chart comparing Artificial Analysis Intelligence Index v4.0 scores: Kimi K2.6 at 53.9, GLM5.1 at 51.4, and GPT-3.5 Turbo at 9.0.
Bar chart comparing Artificial Analysis Intelligence Index v4.0 scores: Kimi K2.6 at 53.9, GLM5.1 at 51.4, and GPT-3.5 Turbo at 9.0.
Bar chart comparing Artificial Analysis Intelligence Index v4.0 scores: Kimi K2.6 at 53.9, GLM5.1 at 51.4, and GPT-3.5 Turbo at 9.0.
Frontier Models

Access all of the smartest models.

ai& runs the frontier open-weight models Kimi, DeepSeek, GLM outperforming proprietary models. For enterprise clients, ai& post-trains on top of these for domain-specific performance.

Heterogeneous Compute

More efficient,
lower costs.

More efficient, lower costs.

Run on heterogeneous Nvidia and AMD hardware, efficient AI serving through at scale disaggregation and optimization to maximising utilisation and throughput, and passing the savings directly to you.

Hardware diagram: inference pipeline using AMD MI355X and NVIDIA B300 chips across prefill and decode stages, producing 154 tokens per second output.
Hardware diagram: inference pipeline using AMD MI355X and NVIDIA B300 chips across prefill and decode stages, producing 154 tokens per second output.
Hardware diagram: inference pipeline using AMD MI355X and NVIDIA B300 chips across prefill and decode stages, producing 154 tokens per second output.
Hardware diagram: inference pipeline using AMD MI355X and NVIDIA B300 chips across prefill and decode stages, producing 154 tokens per second output.
Model Pricing

Priced for production scale.

Industry-leading cost per token, with no contracts required. Run at any scale and pay only for what you use.

Model Name

Input Token Pricing / 1M

Output Token Pricing / 1M

Context Window

Qwen3.6-27B

$0.32

$0.20 (Cached)

$3.20

262.1K

Qwen3.8-27B

$0.40

$0.20 (Cached)

$3.00

262.1K

gemma-4-31B-it

$0.20

$0.05 (Cached)

$0.50

262K

Kimi K2.7 Code

$0.65

$0.20 (Cached)

$3.50

262.1K

Kimi K3

$3.00

$0.50 (Cached)

$12.50

1M

GLM-5.2

$0.80

$0.20 (Cached)

$3.50

1M

GLM-5.3 Flash

$0.15

$0.03 (Cached)

$0.50

1M

GLM-5.3

$1.00

$0.30 (Cached)

$4.00

1M

DeepSeek V4 Flash

$0.15

$0.08 (Cached)

$0.25

1M

DeepSeek V4.1 Flash

$0.30

$0.02 (Cached)

$0.60

1M

DeepSeek V4 Pro

$1.00

$0.25 (Cached)

$2.50

1M

Motif 3

$0.50

$0.20 (Cached)

$2.00

262.1k

gpt-oss-120b

$0.15

$0.08 (Cached)

$0.60

131K

Features

Production ready
from day one.

Every request processed in Japan, on infrastructure that scales automatically with your usage and easily drops into your existing OpenAI and Anthropic compatible tools.

Hosted In Japan

All inference runs on ai&'s Japan-based infrastructure purpose-built for the lowest possible latency across Asia-Pacific and beyond.

Data Resident

Your data stays in Japan. We never use your inputs or outputs for model training, and we comply fully with Japanese data residency requirements.

Scales With You

From zero to at scale concurrent requests in seconds. No cold starts, no capacity planning, just elasticity that matches your traffic.

OpenAI Compatible

A true drop-in replacement. Switch to ai& by changing one line of code, streaming, tool calling, and structured outputs all work out of the box.

Get Started

Start running inference in Japan.

Start running inference in Japan.

inference

Pay as you go

The fastest way to run the world's most capable AI models. Drop in with one line change and start generating in minutes.

Frontier Model Access
OpenAI Compatible
Zero Data Retention

Run Inference

inference

Pay as you go

The fastest way to run the world's most capable AI models. Drop in with one line change and start generating in minutes.

Frontier Model Access
OpenAI Compatible
Zero Data Retention

Run Inference

inference

Pay as you go

The fastest way to run the world's most capable AI models. Drop in with one line change and start generating in minutes.

Frontier Model Access
OpenAI Compatible
Zero Data Retention

Run Inference

Enterprise

Custom solutions

For teams that need everything and more. Dedicated infrastructure, priority support, and terms built around your requirements.

Reserved Compute
Custom SLA
Advanced Security

Contact Us

Enterprise

Custom solutions

For teams that need everything and more. Dedicated infrastructure, priority support, and terms built around your requirements.

Reserved Compute
Custom SLA
Advanced Security

Contact Us

Enterprise

Custom solutions

For teams that need everything and more. Dedicated infrastructure, priority support, and terms built around your requirements.

Reserved Compute
Custom SLA
Advanced Security

Contact Us