OneTriangle GTM brain

OneTriangle provides optimized open-model inference to AI teams running long-context workloads.

category

LLM inference optimization

market profile

  • text: Long prompts make prefill an important contributor to latency and inference spend.
  • text: Hosted inference competes with both managed model APIs and internally operated GPU serving.

personas

  • title: Head of AI infrastructure
    pains:
    • Long context drives slow time to first token and expensive GPU usage
    • Must evaluate whether a new serving path preserves model quality
  • title: ML platform engineer
    pains:
    • Operating open-weight models consumes GPU and serving engineering time
    • Integrating a new endpoint must fit an existing application

icp

  • text: AI application companies with production LLM traffic and long prompts or agentic workflows are a strong fit; size is not specified.
  • text: Teams deploying open-weight Llama, Qwen, Mistral, Gemma or DeepSeek models on managed or dedicated GPU infrastructure.

message pillars

  • text: Reduce latency and cost through cross-model KV cache transfer.
  • text: Use smaller models for prefill and larger models for decoding.
  • text: Avoid managing the GPU serving stack internally.

objections

  • objection: Will cross-model cache transfer degrade answer quality?
    response: Ask for held-out quality results and model-pair benchmarks; the company says ordinary prefill is the fallback when a pair fails its gates.
  • objection: Will moving inference require major application changes?
    response: Validate the endpoint and deployment requirements in a pilot; the site describes a one-line configuration change for serviced models.

competitors

  • name: Together AI
    why: Offers managed open-model inference and dedicated endpoints.
  • name: Fireworks AI
    why: Managed inference platform for open models and production workloads.

influencers

  • name: Hannah Chung
    url: https://www.linkedin.com/posts/hannahchung1_onetriangle-yc-s26-is-building-the-cheapest-activity-7496242168534163458-rmuo
    why: CEO discussing inference and the company launch.

keywords

  • LLM inference optimization
  • time to first token
  • KV cache transfer
  • long context inference
  • GPU inference cost
  • open model serving

voice

  • technical
  • benchmark-led
  • direct

signals

  • text: A team is deploying long-context agents and complains about first-token latency or GPU spend.
  • text: An organization is migrating to open-weight models or expanding dedicated GPU endpoints.

stack

  • Llama
  • Qwen
  • Mistral
  • Gemma
  • DeepSeek

pricing

model: unknown

motion

sales_led

channels

  • LinkedIn
  • X
  • technical research
  • demo

confidence

medium

Founders

  • name: Hannah Chung
    title: Founder/CEO
    role: CEO
    bio: MIT CS & Econ → @Virtu Financial → Fastest Inference @OneTriangle
  • name: Medha Venkatapathy
    title: Founder/CTO
    role: CTO
    bio: Hi! I studied physics + cs at mit, and we're building the cheapest/fastest inference with KV cache transferability. In my free time, I like to bike and get lost.

Back to YC companies