Midas reports BEAM memory benchmark with zero LLM calls

A Reddit user said they ran their first BEAM benchmark with Midas. The post says Midas reached 0.56 recall@k on BEAM 100K and 0.51 on BEAM 500K. The writer says this used 0 LLM calls, $0 API spend, and 0 data egress. They said 1M and 10M tiers are next.

Key points

  • Midas was tested on the BEAM benchmark, according to the post.
  • The reported BEAM 100K score was 0.56 recall@k.
  • The reported BEAM 500K score was 0.51 recall@k.
  • The writer says the run used no LLM calls, no API spend, and no data egress.
  • The next planned tests are the 1M and 10M tiers.

Quick term guide

BEAM benchmark
A test for how well an AI agent can remember and find information over a long context or memory store.
benchmark
A test used to compare speed, quality, or cost.
API spend
Money paid to use an outside software or AI service through its API.
AI agents
AI agents are AI tools that can carry out steps toward a goal, not just answer once.
model call
One request sent to an AI model to get an answer.
local-first
An app design where your data is mainly stored and controlled on your own device.
long-term memory
The capability of an AI to store and retrieve information from previous interactions over extended periods.
token cost
The money or usage spent when sending text to an AI model and getting text back.
Read original