Moonshot AI
Available
Kimi K3.1 Pro
For complex coding, agent and research work, with major gains in long-horizon coding.
- Context
- 1M tokens
- Input
- Text & image
- Tool calling
- Supported
Now in preview Kimi K3.1 Pro available
Inference for AI builders
Run frontier open-weight models with performance engineered for every request, from a single prompt to agents that run for hours.
Built for the way agents work
Long agent sessions are hard to serve. Here's how we keep them fast.
-
Over 99% of reusable agent context served from cache*
-
OpenAI compatible, so your code works as it is
-
Pay per token, with cached input for less
-
Up to 10x faster time to first token on long-context work
*In internal testing, from one recorded agent session with Kimi K3.1 Pro. Reusable context is everything the agent sent in earlier calls.
The right model for every task
Leading open-weight models, from the most capable to the fastest.
z.AI
Coming soon
GLM 5.3
Strong coding and agent performance at a lower cost, for long-running tasks.
- Context
- 1M tokens
- Input
- Text
- Tool calling
- Supported
DeepSeek
Coming soon
V4.1 Flash
A fast, low cost model for high-volume, everyday tasks.
- Context
- 1M tokens
- Input
- Text & image
- Tool calling
- Supported
Engineered for every request
Long agent sessions are hard to serve. Here's how we keep them fast.
Hardware and software, designed together
We run our own GPUs, with serving software we build and tune for them, so every layer is ours to optimize. In six weeks, that let us fit 16 times more agent sessions on each GPU.
Context reused, not reprocessed
Agents resend their history every step. We cache it, so it's never processed twice.
Long prompts don’t hold others up
Reading prompts and writing answers run on separate GPUs, so one long prompt doesn’t stall the rest.
From coding agents to everyday features
Coding agents
Read codebases, make changes, run tests and revise.
Research and analysis
Whole reports and datasets in a single request.
Agent workflows
Long-running agents that plan, act and check their work.
Product features
Chat, summaries and classification in your product.
Agent session Example: Coding agent, Kimi K3.1 Pro on Verda Inference
Recorded coding agent session with 11 model calls, replayed at an accelerated pace. Context grows from 9.9K to 27.3K tokens. 92% of input tokens are served from cache. Median time to first token is 148 milliseconds; the maximum is 161 milliseconds.
| Call | Tool | Action | Context tokens | Cached input | Time to first token |
|---|---|---|---|---|---|
| 01 | search | List repo, find router and tier code | 9,925 | 98% | 144 milliseconds |
| 02 | read | List global-router, read README | 11,675 | 85% | 148 milliseconds |
| 03 | search | Grep tier in gateway upstream | 14,731 | 80% | 149 milliseconds |
| 04 | read | Read admission.rs, list tmp/, design/ | 16,417 | 90% | 148 milliseconds |
| 05 | read | Read existing per-tier doc | 18,032 | 91% | 161 milliseconds |
| 06 | read | Read doc lines 100–280 | 20,470 | 88% | 147 milliseconds |
| 07 | read | Read doc lines 280–415 | 22,666 | 90% | 54 milliseconds |
| 08 | search | Verify queue keys and defaults | 24,545 | 92% | 57 milliseconds |
| 09 | read | Verify transfer(), gateway policy | 25,930 | 95% | 159 milliseconds |
| 10 | search | Verify headers and mutex layout | 26,707 | 97% | 158 milliseconds |
| 11 | reply | Report: doc verified, no changes | 27,308 | 98% | 150 milliseconds |
More ways to run inference
Plan production capacity with our team, or bring your own model.
Production workloads
Running production workloads? Plan them with us.
We'll size capacity, limits and support around your launch, including during the preview.
-
Capacity sized to your launch. Limits planned around your traffic, so growth doesn't hit a default ceiling.
-
One stack, no handoffs. We run the GPUs, the serving software and the data centers ourselves.
-
A direct line to our engineers. The people who run the platform, not a ticket queue.
Bring your own model
Have your own model?
Run your own or fine-tuned models on Verda serverless containers, with GPUs that scale automatically for inference and batch jobs.
-
Any model, any engine. Run SGLang, vLLM or your own server.
-
Scales with demand. Up with traffic, down when idle.
-
One cloud. The same console and billing as Verda Inference.
Available now
Start building with Verda Inference
Make your first call in minutes.