Open-weight models.
Production-grade inference.
SonarCompute is an independent inference provider. We serve open-weight language models through an OpenAI-compatible API, with transparent pricing, predictable latency and zero prompt retention.
Served today
The models we run in production. The list grows as new capacity comes online.
Kimi K2.6
Open-weight MoE model with 1T total parameters and 32B activated. Built for long-horizon agentic tasks: multi-step coding, tool use and autonomous workflows.
MiniMax M2.7
Open-weight model tuned for software-engineering work: agentic coding, log analysis, troubleshooting and refactoring.
Built for routers & platforms
Made to plug into LLM aggregators and high-volume production apps.
Distributed GPU fleet
Multi-region serving with automatic failover. Capacity follows demand; hot models stay warm.
Transparent metering
Exact token counts returned on every response, streaming included. What you see is what gets billed.
Privacy by default
No prompt logging, no training on customer data, zero retention of completions. Read the policy →
Drop-in OpenAI-compatible
- /chat/completions with full streaming support (SSE)
- /models endpoint with live pricing & context metadata
- Usage accounting in every response, stream and non-stream
- Graceful backpressure — early 429s under load, never silent queues
# One-line switch from OpenAI from openai import OpenAI client = OpenAI( base_url="https://api.sonarcompute.com/v1", api_key=os.environ["SONARCOMPUTE_KEY"]) client.chat.completions.create( model="moonshotai/kimi-k2.6", stream=True, messages=[{"role": "user", "content": "Refactor this service…"}])
Route traffic to SonarCompute
Aggregators, platforms and high-volume teams: we onboard new integrations within days. Monthly invoicing in USD, custom SLAs on request.
hello@sonarcompute.com