Comet (Opik)

Researched 2026-06-28. Primary category: llm-observability.

One-liner — Established MLOps vendor (Comet) whose open-source Opik product brings LLM tracing, evaluation, and monitoring.

What it does — Opik (launched 2024-09-17, Apache-2.0 OSS) records every step an LLM/agent app takes (tracing), runs built-in eval metrics and LLM judges (hallucination, factuality), supports “model unit testing” and scoring, and ships production dashboards. It sits on top of Comet’s longer-standing platform for ML experiment management, model management, and production monitoring — so a buyer gets both classic MLOps and LLM observability from one vendor.

Where it sits in the stack — The llm-observability layer of the model/prompt tier. Dev/CI- and ops-time tracing and eval; not an inline runtime guardrail — it does not itself screen prompts or block data flows. Complements ai-red-teaming and ai-runtime-security.

Deployment & architecture — SDK-based instrumentation; Opik self-hostable via OSS (GitHub, ~19k+ stars, 40+ framework/provider integrations) or run on Comet’s managed/enterprise platform with collaboration, user management, and security controls. Pairs naturally with Comet’s experiment-tracking for teams that train/fine-tune as well as call APIs.

Positioning & differentiators — Differentiated by the MLOps lineage: useful if you also do traditional ML and want one platform spanning experiments → models → LLM evals. Opik’s OSS + permissive license puts it alongside langfuse and arize-phoenix on the open-source axis; versus eval-first braintrust and proxy-based helicone it is SDK-based and observability+eval-oriented. The nearest direct comparison is W&B Weave (weights-and-biases) — both extend an incumbent ML-experiment platform into LLM observability.

Ownership, funding & M&A — Independent, VC-backed. Comet ML founded 2017; HQ New York; ~$70M raised (Series B led by OpenView; earlier seed rounds led by Trilogy Equity Partners). 150+ enterprise customers (Netflix, Uber, Cisco, Etsy, Zappos). No M&A. Confidence: high.

CTO / hedge-fund lens — Day-1 for a quant/ML-heavy fund that already trains models and wants LLM observability without adding a separate vendor — the unified MLOps + LLM story is the draw. Opik OSS self-hosting keeps traces (potential MNPI) in-boundary. Smaller/older funding base than the hottest LLM-native startups, but the company is well-established with real enterprise traction. Not a governance/SR 11-7 system itself — pair with ai-governance-platform.

Competitors / alternatives — langfuse, arize-phoenix, braintrust, langsmith, helicone, datadog, weights-and-biases.

Open questions / to verify

• Exact latest funding round size/date (only ”~$70M total, Series B led by OpenView” confirmed).
• Opik OSS vs Comet-managed feature boundary.

Sources

• Comet Launches Opik, an Open Source LLM Evaluation Platform (Comet press release) — fetched 2026-06-28 — supports: Opik launch 2024-09-17, OSS, Comet founding/HQ/funding/customers, core MLOps platform; confidence: high.

History

• [2026-06-28] Stub created from seed registry.
• [2026-06-28] Researched; confirmed independent (Comet ML, ~$70M, Series B OpenView), established Opik (OSS LLM eval, launched 2024-09) vs core MLOps platform, filled founding/HQ/deployment. No M&A. ownership_confidence low→high.