
Latent
Latent is a YC Fall 2026 (F26) company. Based in San Francisco, CA, USA. Team of 2.
About Latent
Summary
Latent detects hallucinated LLM outputs in real time by reading a model's internal state as it generates each answer, before any response reaches an end user. Teams deploying LLMs in finance, legal, healthcare, and customer support get a prioritized review queue of flagged answers, each with a plain explanation, replacing the manual spot-checking or expensive judge-model pipelines those teams would otherwise run. The product sits inside the customer's own deployment, so no output leaves their environment to be verified.
Business model
A free tier plus paid plans, with a call-to-book sales entry point.
Go-to-market
B2B teams running LLMs in production are the buyers; they can start via a self-serve free tier or book a call for a guided onboarding.
From the Y Combinator listing
Automated post-training for LLMs
Latent automates post-training for LLMs. It reads signals inside a model as it works and labels every answer and agent step in milliseconds, so teams improve their models on their own production traffic instead of paying for human labeling. The same signal works in real time. Latent lets an output through, holds it, re-asks a stronger model, or sends it to review with the reason attached. Reading the model is over 660x cheaper than running an LLM judge, so it can check everything. One voice AI company using Latent finds 98% of its agent’s failures by reviewing only the riskiest 43% of answers, and in testing Latent stopped 99% of rule-breaking agents before they acted, against 69% for a frontier LLM judge. It works with self-hosted and API models like GPT, Claude, and Gemini, runs hosted or fully in your own cloud, and installs with one line of code.
Facts
- B2B / B2B
- San Francisco, CA, USA
- 2
- Active
- Early
Founders (1)
- Kunal Ranjan
Score every batch against your own thesis
GPAgent's sourcing agent reads every company in a batch, scores each one against your fund's thesis in a private workspace, and lines up the founders worth meeting before Demo Day. You decide who to contact. Your scores, notes, and pipeline never appear on these pages.
How Accelerator intelligence works