Comet - Licensing only

What do you like best about the product?

The platform is best for when we are working in a team on a ML Model. It has a good collaboration features, reports and panels which can be share among the team members. Free tier for individuals and academics person which suits students build portfolio.

What do you dislike about the product?

It is mainly focus on Python, it supports other languages very less. It track and monitor the ML Models but does not run your pipelines like Cubeflow or airflow and the team and enterprise addition of this platform is very expensive as we compared to open source options.

What problems is the product solving and how is that benefiting you?

The machine learning work is experimental and messy, you try dozens of code to run with different hyperparameter dataset codeversion and architecture but after we use format we get to know that it is an experimental tracking and ML Lifecycle program that records everything about each run so the results are comparable, sharable and reproducible.

What do you like best about the product?

I explored Comet's experiment observability capabilities through its Opik tracing workflow. I found it useful to see function inputs and outputs recorded in a centralized dashboard, making it easier to inspect execution results and understand what happened during a run. The trace details were straightforward to navigate, and having the input, output, and execution duration visible together was helpful for debugging a small test workflow. For someone learning to monitor AI application behavior, this kind of visibility can be a useful starting point. My evaluation was limited to a small hands-on test, so I have not assessed the full experiment-tracking feature set or its value in a production environment.

What do you dislike about the product?

My evaluation was limited to a small tracing workflow, so I haven't explored enough of Comet.ml to assess all its limitations. During setup, the initial configuration required a few steps, and clearer guidance around API key setup and project configuration would make onboarding smoother for new users. A more beginner-friendly quickstart with troubleshooting tips and a simple example could help users get their first trace working faster. I would need more hands-on experience with the broader platform before commenting on its performance, pricing, or advanced features.

What problems is the product solving and how is that benefiting you?

I explored Opik's tracing workflow to understand how application execution can be monitored and debugged. For my small test, it recorded a function's input and output in a centralized dashboard, making it easier to inspect the result without relying only on terminal output. This was useful for learning how tracing can help developers investigate application behavior and identify issues. My evaluation was limited to a test function, so I haven't measured business impact or used Comet.ml in a production workflow.

What do you like best about the product?

Comet is the newest open-source tool for tracing and evaluating ML LLM applications, reflecting the shift toward LLM work. Check their site for the current feature set, as this part is evolving very quickly. The main thing about this platform is that it is 3-tier for individuals and academics, which suits students building portfolios.

What do you dislike about the product?

The most hated part is that it has a vendor dependency because the full-fledged product is hosted as SaaS, so your experiment history lives on their platform, and self-hosting is typically an enterprise feature. It is Python-centric and supports other languages less well.

What problems is the product solving and how is that benefiting you?

When you work on an LLM model, the results will be scattered across notebooks, terminal output, and spreadsheets, causing much confusion. But when we started using Comet, most of the time, whenever we ran our LLM models, it recorded every single output in the same format. Comet can put them side by side, allowing you to sort by accuracy, overlay loss curves, and hyperparameter changes.

What do you like best about the product?

it has flexible deployment on cloud on premises for enterprise with data control needs you can collaborate with your team and share project panel and report with each other.

What do you dislike about the product?

Basic logging is easy but the control panel artifacts and monitoring takes time to master which causes overhead for small projects.

What problems is the product solving and how is that benefiting you?

The model works in testing but in the real world the data changes every day and accuracy drops quietly because ML experiments generate a lot of info and humans track it very badly by hand. Now Comet acts like a lab notebook that writes itself, recording automatically and storing everything in a central place. It lets us compare and analyze the data and helps us track the model beyond training.

What do you like best about the product?

The most liked part about this platform is instead of forcing HTTP requests every time a metric is calculated, which would cripple the GPU training performance, the SDK pushes raw metrics into an in-memory queue. It automatically batches the payload and flushes them to the Comet server over a persistent WebSocket connection.

What do you dislike about the product?

The most disliked part about this platform is it has advanced features, a common frustration among ML engineers is the occasional drift between what can be done by the Python SDK versus what functions in the browser UI.

What problems is the product solving and how is that benefiting you?

A training job run for 14 hours on a cloud GPU cluster and suddenly crashes and underperforms. Without live monitoring, the engineer building that ML model has no idea if it ran out of VRAM, overheated, or suffered from exploding gradients. Here, Comet comes in and streams log real-time hardware metrics alongside training metrics. If a model fails, you can see the exact moment the hardware or math broke down.

What do you like best about the product?

The most liked part is that it automatically generates clean graphs, confusion matrices, histogram plots, and system utilization charts out of the box with minimal boilerplate code.

What do you dislike about the product?

The most disliked part is that it has a large feature set, which can feel overwhelming or like overkill when you are building something simple. For first-time users, it feels very tricky.

What problems is the product solving and how is that benefiting you?

This platform is solving the chaos that comes with building and working on an ML lifecycle. Before this platform existed, managing ML experiments looked a lot like software development before Git, relying on messy spreadsheets, manual notes, and lost configurations.

What do you like best about the product?

It is very easy to setup you can setup it in very few lines of code and you can autolog in a minute it work across many python ml libraries like skitlearn and the code environment and parameter are saved automatically saved after one run of ml model.

What do you dislike about the product?

It is mainly python centric it supports less other languages like ruby kotlin and java and the cost of team and enterprise edition camn get very expensive as compared to any open source option like github.

What problems is the product solving and how is that benefiting you?

Building an ml in this day to day life is not same as writing normal software, the same code give same output but in ml the result depends upon the code , data which you upload what setting you setup some randomness and depends upon which environment you are working and you can change all of these constantly while experimenting the ai but when we use this comet ml platform, it make a proper environment for a team so it has same data setting in an environment which improves the quality of ml model as compared to a normal software

What do you like best about the product?

The best part is when we build a model in a team, it helps us check which team member is working on which part of the model, reducing time and duplication.

What do you dislike about the product?

The most disliked part of this software/platform is that we can't predict the price and the amount of credits that will be used on a project, leading to high usage and platform fees that are not easily affordable by a normal organization.

What problems is the product solving and how is that benefiting you?

When working on traditional software to create a model like GitHub, we can't identify which part of the work is being done by which member, leading to long hours and reducing our team's potential. Comet ML solves the problem of duplication among team members.

What do you like best about the product?

It helps us in data science projects to track, compare, visualize, and optimize machine learning experiments and LLM workflows in one place, and you can integrate Comet into your training script by adding just a few lines of code.

What do you dislike about the product?

It has paid tiers and usage fees in billing, such as extra usage of Opik or per user seat licenses, which increase the cost to use it and rise with an increase in team members and growing data.

What problems is the product solving and how is that benefiting you?

In engineering teams, team members often work individually on their local machines without a centralized hub, causing duplication of work when two members do the same task, leading to time loss. However, when we use Comet.ml, it helps team members track who is working on which project and then work accordingly.

What do you like best about the product?

Before Comet, the “best” model lived in someone’s memory or an old notebook name. What I like is that every serious run leaves a record—metrics, parameters, and context—in one place, so we can find what actually worked instead of re-running guesses. Comparing a few candidates takes minutes, and handing a result to someone else does not depend on them recreating my local setup.

What do you dislike about the product?

If the team logs inconsistently, the workspace gets noisy. You still need light discipline on what to track and how to name projects. It will not fix a messy process by itself.

What problems is the product solving and how is that benefiting you?

It solves the lost-experiment problem: results scattered across machines and chat threads. The benefit is faster decisions and less repeated work, because we can see prior runs and build on them instead of starting from scratch each time.