Neptune.ai

Name: Neptune.ai
Brand: Neptune.ai
Rating: 4.6 (54 reviews)

Experiment tracker purpose-built for foundation model training.

Est 2016|Valuation $300M|Raised $78.7M|33 Ref Customers

LinkedIn 41K|Web traffic 10K|FTEs 115|HQ Palo Alto, California, US

MarketsData Science & Machine Learning Platforms Machine Learning Ops (MLOps)

Published Oct 9, 2026

01Common complaints

Given the research date (January 26, 2026) and the vendor's announced discontinuation of standalone services, consider avoiding Neptune for the following situations: 1) Any net-new purchase where the intent is to use Neptune beyond March 5, 2026 - The vendor has announced that standalone services will be discontinued and hosted data will need to be exported by the stated date. 2) Organizations that cannot tolerate near-term migration risk - If experiment history, record retention, or reproducibility are compliance-critical, the fixed migration timeline imposes significant operational and compliance risk. 3) Teams that require easy external sharing or view-only access for stakeholders - Reviews report licensing friction when sharing results with users who do not hold a paid license. 4) Budget-sensitive teams and smaller ML groups - Published tiered pricing is relatively high compared with some alternatives and may not fit low-cost or early-stage teams. 5) Organizations that require active feature development or platform expansion (LLM evaluation, tracing, model registry depth) - The vendor has stated no new features will be released during the transition period.

Vendor-acknowledged limitations / lifecycle constraints

Services are scheduled to be discontinued on March 5, 2026; hosted application and APIs will go offline unless customers export data.
No new features are planned during the transition; only stability and security fixes are expected.

Technical/product limitations surfaced in reviews

Dashboard layout control and resizing behavior are limited.
Grouping runs within projects is reported to be difficult at scale.
Offline logging synchronization has differences compared with asynchronous behavior.
Sharing and collaboration with non-licensed users can be constrained.

Roadmap

The vendor has effectively frozen the public roadmap due to the acquisition and planned discontinuation of standalone services.

02Why they buy

Distinctive capabilities supported by evidence

Performance at scale: multiple reviewers report the UI and backend can handle large numbers of experiments and metric volumes.
Foundation-model observability: vendor materials and the OpenAI relationship highlight per-layer metrics and multi-run analysis for large-scale training.
Self-hosting availability: the product offers a self-hosted deployment option, which supports on-premises or private cloud installs.

Areas where competitors have advantages

Long-term viability and roadmap: alternatives such as W&B, ClearML, Comet, and MLflow do not have announced discontinuations in the materials reviewed.
Broader platform coverage: some competitors include pipeline orchestration, HPO, registry/lineage, and production monitoring in their offerings.

03Field reports

Review platform summary (G2)

- G2 rating: very highly rated on G2, though review volume is limited. Representative positive quotes from G2 (excerpts): - "We use Neptune consistently to track hundreds of different experiments… It provides a consistent structure that ensures all our data is reliably and intuitively organized." (Senior ML Engineer) - "Easy to set up, plenty of integration, useful collaboration tool." (Staff AI Engineer) - "The API is simple to set up and requires minimal code to track experiments." (Data Scientist) Representative criticisms from G2 (excerpts): - "Offline logging behavior may diverge from the standard asynchronous behavior…" (Senior ML Engineer) - "The only downside for me is the lack of customizability in the front-end… build tailor-made dashboards…" - "Unable to share results with anyone without the license." (Staff AI Engineer) - "Dashboard visualizations automatically resize… Would love if there was a way to fix the size…" (Data Scientist) - "Grouping runs within a project is difficult, this makes usage for large teams challenging" (ML Engineer) Interpretation: Reviews commonly note ease of setup, responsive UI, and support quality, while complaints focus on dashboard customization, sharing constraints, grouping/organization at scale, and some workflow edge cases (offline mode, non-monotonic steps).

Capterra

A third-party analysis indicates that Capterra shows no vendor reviews, which is consistent with limited coverage on some review platforms.

Community / social proof

Public community discussion was limited in the reviewed crawls, but Neptune maintains an active GitHub presence and public client libraries.

Employee sentiment (Glassdoor)

Glassdoor shows limited review volume and includes negative commentary among the available entries. The sample size is small and may not be representative, but reported employee sentiment can be a data point for considerations such as support continuity during a transition.

Vendor testimonials (not independently verified)

The vendor's homepage includes testimonials that assert reliability and UI performance (for example, statements such as "never seen any outage" and "it's fast, it's reliable"). These are vendor-provided statements and were not independently verified; audited uptime or performance reports were not found in the public materials reviewed.

OpenAI relationship as a signal

The vendor states that OpenAI used Neptune’s tooling for monitoring and debugging large-scale training. Public reporting on the acquisition corroborates a working relationship between the organizations. Critical evaluation: - The vendor's engagement with a leading frontier-scale organization indicates relevance for frontier-scale workloads, but it does not guarantee suitability for smaller teams or different operational contexts. - The acquisition and planned discontinuation of standalone services limits Neptune's availability as an independent vendor for external buyers.

04Best fit

Best-fit organizations (historical)

Company size: mid-market to enterprise ML organizations that run many experiments and require team-wide standardization.
Workloads: long-running training jobs, distributed training, heavy metric logging (including per-layer metrics).
Industries: AI-first companies, biotech, and large enterprises training proprietary models.

Personas

Decision makers: Head of ML, Director of AI, ML platform lead, research engineering lead
End users: research scientists, ML engineers, MLOps/platform engineers

Qualification criteria (good fit)

Requirements include fast search and visualization across very large time series and tables.
Teams that benefit from standardized metadata conventions and structured comparison workflows.
Organizations that require self-hosting for security or compliance, understanding that a near-term migration plan is required given the vendor's announced timeline.

Disqualifiers

Organizations that require a vendor with a multi-year public roadmap and publicly documented product availability commitments.
Organizations that require a broader end-to-end MLOps platform (HPO, orchestration, production monitoring) in a single product.

05The pitch

What the product does (core value proposition)

Neptune.ai is a training observability and experiment-tracking platform for logging, visualizing, searching, and comparing large volumes of model-training metadata (metrics, parameters, artifacts). The product emphasizes foundation-model-scale training, including per-layer metrics such as losses, gradients, and activations. The platform is designed to render large charts and tables with minimal downsampling or UI lag, and is available as a hosted (SaaS) service and as a self-hosted deployment option. A material product lifecycle event is underway: the vendor has entered a definitive agreement to be acquired and the vendor’s standalone services will be discontinued in the near term. This change should be a primary consideration in procurement decisions.

Target market and positioning

- Primary target (historically): ML teams needing experiment tracking (data scientists / ML engineers), with differentiation around handling large-scale metadata and UI responsiveness. - Current vendor positioning: described as an experiment tracker focused on foundation models, with attention to per-layer metrics and large-scale training observability. - Segments referenced by vendor: research labs, foundation model builders, and enterprise teams that require self-hosting for security or compliance.

Get a comprehensive analysis of Neptune.ai including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →

Published Oct 9, 2026

01Common complaints

Given the research date (January 26, 2026) and the vendor's announced discontinuation of standalone services, consider avoiding Neptune for the following situations:

Any net-new purchase where the intent is to use Neptune beyond March 5, 2026

The vendor has announced that standalone services will be discontinued and hosted data will need to be exported by the stated date.

Organizations that cannot tolerate near-term migration risk

If experiment history, record retention, or reproducibility are compliance-critical, the fixed migration timeline imposes significant operational and compliance risk.

Teams that require easy external sharing or view-only access for stakeholders

Reviews report licensing friction when sharing results with users who do not hold a paid license.

Budget-sensitive teams and smaller ML groups

Published tiered pricing is relatively high compared with some alternatives and may not fit low-cost or early-stage teams.

Organizations that require active feature development or platform expansion (LLM evaluation, tracing, model registry depth)

The vendor has stated no new features will be released during the transition period.

Vendor-acknowledged limitations / lifecycle constraints

Services are scheduled to be discontinued on March 5, 2026; hosted application and APIs will go offline unless customers export data.
No new features are planned during the transition; only stability and security fixes are expected.

Technical/product limitations surfaced in reviews

Dashboard layout control and resizing behavior are limited.
Grouping runs within projects is reported to be difficult at scale.
Offline logging synchronization has differences compared with asynchronous behavior.
Sharing and collaboration with non-licensed users can be constrained.

Roadmap

The vendor has effectively frozen the public roadmap due to the acquisition and planned discontinuation of standalone services.

02Why they buy

Distinctive capabilities supported by evidence

Performance at scale: multiple reviewers report the UI and backend can handle large numbers of experiments and metric volumes.
Foundation-model observability: vendor materials and the OpenAI relationship highlight per-layer metrics and multi-run analysis for large-scale training.
Self-hosting availability: the product offers a self-hosted deployment option, which supports on-premises or private cloud installs.

Areas where competitors have advantages

Long-term viability and roadmap: alternatives such as W&B, ClearML, Comet, and MLflow do not have announced discontinuations in the materials reviewed.
Broader platform coverage: some competitors include pipeline orchestration, HPO, registry/lineage, and production monitoring in their offerings.

03Field reports

Review platform summary (G2)

G2 rating: very highly rated on G2, though review volume is limited.

Representative positive quotes from G2 (excerpts):

"We use Neptune consistently to track hundreds of different experiments… It provides a consistent structure that ensures all our data is reliably and intuitively organized." (Senior ML Engineer)
"Easy to set up, plenty of integration, useful collaboration tool." (Staff AI Engineer)
"The API is simple to set up and requires minimal code to track experiments." (Data Scientist)

Representative criticisms from G2 (excerpts):

"Offline logging behavior may diverge from the standard asynchronous behavior…" (Senior ML Engineer)
"The only downside for me is the lack of customizability in the front-end… build tailor-made dashboards…"
"Unable to share results with anyone without the license." (Staff AI Engineer)
"Dashboard visualizations automatically resize… Would love if there was a way to fix the size…" (Data Scientist)
"Grouping runs within a project is difficult, this makes usage for large teams challenging" (ML Engineer)

Interpretation: Reviews commonly note ease of setup, responsive UI, and support quality, while complaints focus on dashboard customization, sharing constraints, grouping/organization at scale, and some workflow edge cases (offline mode, non-monotonic steps).

Capterra

A third-party analysis indicates that Capterra shows no vendor reviews, which is consistent with limited coverage on some review platforms.

Community / social proof

Public community discussion was limited in the reviewed crawls, but Neptune maintains an active GitHub presence and public client libraries.

Employee sentiment (Glassdoor)

Glassdoor shows limited review volume and includes negative commentary among the available entries. The sample size is small and may not be representative, but reported employee sentiment can be a data point for considerations such as support continuity during a transition.

Vendor testimonials (not independently verified)

OpenAI relationship as a signal

Critical evaluation:

The vendor's engagement with a leading frontier-scale organization indicates relevance for frontier-scale workloads, but it does not guarantee suitability for smaller teams or different operational contexts.
The acquisition and planned discontinuation of standalone services limits Neptune's availability as an independent vendor for external buyers.

04Best fit

Best-fit organizations (historical)

Company size: mid-market to enterprise ML organizations that run many experiments and require team-wide standardization.
Workloads: long-running training jobs, distributed training, heavy metric logging (including per-layer metrics).
Industries: AI-first companies, biotech, and large enterprises training proprietary models.

Personas

Decision makers: Head of ML, Director of AI, ML platform lead, research engineering lead
End users: research scientists, ML engineers, MLOps/platform engineers

Qualification criteria (good fit)

Requirements include fast search and visualization across very large time series and tables.
Teams that benefit from standardized metadata conventions and structured comparison workflows.
Organizations that require self-hosting for security or compliance, understanding that a near-term migration plan is required given the vendor's announced timeline.

Disqualifiers

Organizations that require a vendor with a multi-year public roadmap and publicly documented product availability commitments.
Organizations that require a broader end-to-end MLOps platform (HPO, orchestration, production monitoring) in a single product.

05The pitch

What the product does (core value proposition)

A material product lifecycle event is underway: the vendor has entered a definitive agreement to be acquired and the vendor’s standalone services will be discontinued in the near term. This change should be a primary consideration in procurement decisions.

Target market and positioning

Primary target (historically): ML teams needing experiment tracking (data scientists / ML engineers), with differentiation around handling large-scale metadata and UI responsiveness.
Current vendor positioning: described as an experiment tracker focused on foundation models, with attention to per-layer metrics and large-scale training observability.
Segments referenced by vendor: research labs, foundation model builders, and enterprise teams that require self-hosting for security or compliance.

Get a comprehensive analysis of Neptune.ai including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →