COMET ML

Gideon Mendels and Bar Weber were both working in machine learning and kept hitting the same wall. You train a hundred models over three months.

Then a colleague asks why last week's version was better. Nobody knows.

You can't reproduce it. You lost the config file.

The experiment notes are in someone's Jupyter notebook that no longer runs. They built Comet to solve that exact problem — a logging layer that automatically captures every parameter, metric, and artifact from every training run.

The first version was built in a weekend. Enterprise customers followed quickly.

SaaS. Teams pay for seats and compute quotas to log experiments, track metrics, compare model runs, and manage model registries.

Pricing scales from free for individual developers to six-figure enterprise contracts with private cloud deployment, custom data retention, and SSO. The newer Comet Opik product — LLM evaluation and observability — runs on the same platform and adds usage-based pricing for LLM tracing.

Comet Experiment — real-time experiment tracking, metric logging, parameter capture, and model comparison across training runs. Comet Opik — LLM observability, prompt evaluation, and trace logging for production AI applications.

Comet Model Registry — centralized artifact management and deployment tracking for ML teams.

Bottom-up developer adoption. Individual data scientists start on the free tier, experiment tracking clicks immediately, and they bring it into their team.

Teams convert to paid. Teams convert to enterprise.

It's the Datadog and GitHub motion: get the individual, land the team, expand the contract. The LLM wave added a second surface area — every company building with OpenAI or Anthropic now needs evaluation tooling, and Comet positioned Opik directly there.

Weights and Biases built stronger community momentum first. They attracted top-tier academic ML labs early, published deep integrations with every major framework, and became the default in research circles.

In developer tools, default is almost impossible to displace. Comet has had comparable functionality but has played catch-up on mindshare for years.

Second place in a market dominated by a single standard is genuinely difficult.

Tiger Global, Intel Capital, Trilogy Equity Partners, Felicis Ventures