MarsTrace, kernel tracing as a service, is in early access →

AI-native infrastructure that improves itself.

From silicon to API to applications

MarsCompute serves open models through one API, builds the tools that trace and tune GPU kernels, and ships AI-native products on top: one stack for the industries, enterprises and tasks that run on AI.

Technology

Agents that make the GPUs under AI faster.

Our optimization work is a loop, not a one-off tuning pass. Each turn leaves a faster kernel and a record of what worked, which the next agent reads before it starts. That is recursive self-improvement, applied to AI infrastructure.

1 · Trace 2 · Diagnose 3 · Rewrite 4 · Verify Every turn a faster kernel and a cheaper next turn
  1. Trace

    A timeline of every warp inside the compiled kernel, recorded without changing the kernel.

  2. Diagnose

    A coding agent reads the trace and finds the stage that is waiting, and what it is waiting for.

  3. Rewrite

    The agent changes the kernel: a pipeline stage, a schedule, a parameter.

  4. Verify

    The new kernel is measured and checked for accuracy. What works is kept, and the loop starts again.

3.9×

Sooner to the target

An agent tuning FlashAttention-3 reached its target at iteration 14 with our trace, against 54 with the best existing tracer.

G-Watch, September 2026

+13.3%

FlashAttention-4 on B300

Over the open-source kernel, after an agent studied traces of a closed-source one. 5.2 to 13.3% across shapes.

G-Watch, September 2026

+50.8%

FlashAttention-2 on A100

Best case from an automated run; 25 to 51% at sequence length 1,024 and up with batch size 4 and up.

Report, March 2026

19

GPU architectures

NVIDIA from Volta to Blackwell, and AMD across CDNA and RDNA, traced the same way.

G-Watch, September 2026

Token Factory

Every model through one API

Already using an OpenAI SDK? Change the base URL and the key, and keep the rest of your code.

  • OpenAI-compatible

    Chat completions and streaming in the format your SDK already speaks. To switch models, change the model field.

  • One key, every model

    A single key reaches every text and video model. Limit a key to one model when you want to, and revoke it at any time.

  • Pay only for what you use

    Prepaid credit from $10, charged per request at the published price. A request that fails before answering is not charged.

import os
from openai import OpenAI

client = OpenAI(
  base_url="https://api.marscompute.ai/v1",
  api_key=os.environ["MARSCOMPUTE_API_KEY"],
)

reply = client.chat.completions.create(
  model="qwen3-30b-a3b-instruct-2507",
  messages=[
      {"role": "user", "content": "Count from one to five."},
  ],
  max_tokens=256,
)
print(reply.choices[0].message.content)
Working with teams at Meta 4Paradigm AI Singapore
Customer story

Meta

A customized G-Watch for internal AI workloads

Under a commercial collaboration, MarsCompute provided Meta with a customized G-Watch to help its engineers optimize internal AI workloads.

Read the story
Product
G-Watch, customized
Goal
Optimize internal AI workloads
Why it matters
At hyperscale, small kernel gains are large amounts of compute

Build on infrastructure that keeps getting faster.

Sign up for the API in minutes, or talk to us about your own kernels, hardware and workloads.