Humanloop pricing

About

Humanloop was an enterprise platform for building LLM-powered products — prompt management, evaluation (“evals”), and production observability in one workspace shared by engineers and non-technical domain experts. Founded in 2020 as a University College London spinout and a Y Combinator company, it raised roughly $7.9M in total seed funding across two rounds (Index Ventures, Y Combinator, LocalGlobe, AlbionVC). The early product grew out of human-feedback / data-labeling work for machine learning; as GPT-3 and then ChatGPT created sudden demand for prompt tooling, Humanloop repositioned around prompt versioning, eval reports, and online monitoring for teams shipping LLM features.

In August 2025, Anthropic acqui-hired the Humanloop team — co-founders Raza Habib (CEO), Peter Hayes (CTO) and Jordan Burgess (CPO) plus about a dozen engineers and researchers. Crucially, Anthropic confirmed it did not acquire Humanloop’s IP or assets; it bought the talent. As a result Humanloop shut down its platform rather than handing it to new owners, telling customers in mid-2025 it would wind down and helping them migrate through late 2025.

Everything below is therefore a post-mortem: the pricing described here is historical and indicative, not currently purchasable. The canonical URL now resolves to a “Humanloop is joining Anthropic” notice.

For the most current company status, visit Humanloop.

Pricing summary : How Humanloop’s pricing model worked

Humanloop ran a tiered SaaS model metered on usage, where the meter was logs (later) or datapoints (earlier) rather than tokens. A log was created for every call to a Prompt, Tool, Evaluator or Flow — capturing inputs, outputs, metadata, human feedback, and evaluator judgments. Plans bundled a monthly log/datapoint quota; exceeding it triggered a notification and an upgrade conversation rather than a hard cutoff, and higher log volumes earned volume discounts.

By its final public state the page exposed only two visible options: a free trial (2 members, 50 eval runs, 10K logs/month) and a sales-led Enterprise tier (SSO+SAML, RBAC, VPC/self-hosted deployment add-on, SOC-2 Type II, HIPAA via BAA, EU/US hosting, dedicated support). A startup program for early-stage VC-backed companies sat alongside. In its earlier (2023) life the meter was datapoints (a stored model completion) and the page carried explicit self-serve Starter and Team prices with a −30% annual discount, but those self-serve dollar tiers were removed in 2024 when the meter switched to logs and pricing moved behind “Contact Sales” — see Pricing evolution for the exact historical figures.

What makes this different: Humanloop never charged for model tokens — you brought your own OpenAI/Anthropic API key and paid those providers directly. Humanloop monetized the workflow layer (logging, versioning, evals, monitoring) on a usage meter, which is why “log” — not “token” or “seat” — was the load-bearing pricing unit.

Pricing by product

Sales motions across products: self-serve/PLG for the free trial and (historically) self-serve paid tiers, transitioning to a sales-led Enterprise motion as the company moved upmarket. The platform is now sunset; nothing in this table is currently purchasable.

Hidden costs : What Humanloop users actually paid

The biggest “hidden” line item was the LLM bill itself: Humanloop deliberately did not resell model tokens — you brought your own API keys and paid OpenAI/Anthropic separately. So a team’s true cost was Humanloop’s platform fee plus its underlying model spend, which could dwarf the subscription for high-volume apps.

The second was the log meter. Because every prompt/tool/evaluator/flow call — plus external logs you forwarded and every human/evaluator judgment — counted as a log, eval-heavy or high-traffic workloads consumed quota fast. Overage was handled by an upgrade conversation, not a fixed per-log surcharge on the public page, so real cost scaled with how aggressively a team instrumented and evaluated.

Want to estimate a comparable evals/observability bill? Use the Humanloop pricing calculator to model log-metered costs (historical reference).

Pricing evolution : Humanloop pricing history and shutdown

Cadence

Tracked range: 2020–2025, corroborated by Wayback captures (2023-03 through 2025-07). The live page is now an acquisition/sunset notice; the dollar figures below are historical and no longer purchasable.

Notable changes

• 2020 — Founded as a UCL spinout and YC company; raised ~$7.9M seed across two rounds (Index Ventures, Y Combinator, LocalGlobe, AlbionVC). Early focus on human-feedback / data labeling.
• 2023 (datapoint era) — Self-serve tiers metered on datapoints (a datapoint = one model completion stored + enriched with inputs/feedback/corrections): Starter $100/mo (1k datapoints/mo, then $10 per 1k; 1 seat), Team $1,000/mo “most popular” (100k datapoints/mo, then $4 per 1k; unlimited seats, A/B testing, finetuning), Enterprise custom (RLHF finetuning, hosted models, RBAC). Annual billing carried a −30% discount. (Source: Wayback 2023-03 capture.)
• early 2024 — Simplified to Free (2 members, 1k logs/month, community support) + Enterprise (Talk to sales). The meter switched from datapoints → logs, and the published self-serve dollar tiers were removed. (Source: Wayback 2024-03 capture.)
• late 2024 (General Availability) — Free trial expanded to 2 members / 50 eval runs / 10K logs/month; sales-led Enterprise (SSO+SAML, RBAC, VPC add-on) plus a Startup Program. Repositioned as the “LLM Evals Platform for Enterprises.” (Source: Wayback 2024-12 capture.)
• 2025-08-13 — Anthropic announced it acqui-hired the Humanloop team (three co-founders + ~a dozen staff). Anthropic confirmed no IP/assets transferred.
• 2025 (mid → late) — Humanloop told customers it would shut down ahead of the deal and sunset the platform, assisting with migration. Product no longer purchasable.

What’s unique : Humanloop’s distinctive pricing mechanics

1. The meter was the “log,” not the token. Humanloop priced the workflow layer — every prompt/tool/evaluator/flow call, plus forwarded logs and human/evaluator judgments — instead of reselling model inference. That made the unit economics about observability volume, decoupled from raw LLM cost.

2. Bring-your-own-key, pay providers directly. By design you used your own OpenAI/Anthropic keys and any custom terms with them, and fine-tuned models stayed accessible outside Humanloop. Humanloop never marked up tokens, which kept its pricing legible but meant the platform fee was only part of the real bill.

3. Soft overage, not a hard wall. Exceeding plan limits triggered a notification and an upgrade conversation rather than cutting off service — explicitly framed as “we never want limit overages to disrupt your business,” with volume discounts for higher log counts.

Strengths & weaknesses

Billing UX : Humanloop billing controls and transparency

• Billing controls — Standard plans billed monthly with no long-term commitment; Enterprise could opt for annual billing with additional benefits. Volume discounts available for higher log counts.
• Usage visibility — Plan-limit notifications as a team approached its log/eval quota; overage handled via upgrade conversation rather than automatic surcharge. Data was exportable at any time.
• Payment & deployment — Self-serve monthly subscription for standard plans; Enterprise was custom-quoted (Contact Sales) and could include VPC/self-hosted deployment, EU/US/UK data residency, and dedicated Slack support. Note: as of the 2025 sunset, none of this is purchasable — the page is an acquisition notice.

Strategic wins : Why Humanloop’s pricing decisions worked

1. Pricing the workflow, not the tokens

By metering logs/datapoints and refusing to resell inference, Humanloop kept its value metric aligned with what it actually delivered — evals, versioning, monitoring — rather than competing on token markup. See choosing the right usage metric for why the value metric choice matters.

2. Moving upmarket to enterprise evals

Repositioning as the “LLM Evals Platform for Enterprises,” with VPC/self-hosted, SOC-2 Type II and HIPAA, let it pursue regulated, high-ACV customers — the buyers willing to pay for trustworthy LLM deployment. Related: how AI companies structure pricing.

3. A talent outcome, if not a product one

The lean ~$7.9M raise and a founding team strong enough for Anthropic to acqui-hire is, in startup terms, a “win” — even though the product didn’t survive. It’s a reminder that pricing/monetization success and an acqui-hire exit are different scorecards. See outcome-based pricing trends.

Areas to improve : Gaps in Humanloop’s pricing approach

1. Transparency regressed over time

Burying self-serve dollar tiers behind “Contact Sales” as it moved upmarket made the public page harder to evaluate without a sales call — the opposite of the legible log meter it started with. See bill shock and cost unpredictability.

2. The “real” cost was split across two bills

Because model tokens were paid directly to OpenAI/Anthropic, buyers had to reason about Humanloop’s fee and their provider spend separately. Clearer combined-cost guidance would have reduced sticker surprise for high-volume apps.

3. No standalone continuity after the acqui-hire

The hardest gap is structural, not pricing: an IP-less acqui-hire meant customers faced a forced migration with no in-platform successor. For a tool sold on “trustworthy” production AI, abrupt sunset is the steepest possible switching cost.

Key takeaways

1. Humanloop’s product is gone. Anthropic acqui-hired the team (Aug 13, 2025) without buying the IP, so the platform was sunset — all pricing here is historical/indicative.
2. The meter was the log. It priced workflow volume (prompt/tool/evaluator/flow calls + judgments), not model tokens, decoupling platform cost from inference spend.
3. Free trial → self-serve $ → sales-led Enterprise. Pricing transparency narrowed as it moved upmarket; the last public state showed only a free trial and Contact-Sales Enterprise.
4. Bring-your-own-key kept it honest but split the bill. No token markup, but real cost = subscription + your own provider spend.
5. An acqui-hire is not a pricing endorsement. A lean ~$7.9M raise and a desirable team produced a talent exit, not a durable monetization story — a useful distinction for the LLMOps category.

UBP implications

1. Choose a value metric you control and can defend. “Logs” tied Humanloop’s revenue to instrumented activity it actually provided — a cleaner story than marking up tokens it didn’t. See the introduction to usage-based pricing for how to pick a meter that aligns price with delivered value.
2. Transparency is a feature; losing it has a cost. Pulling published self-serve prices behind sales gates trades evaluability for ACV — a tradeoff LLMOps vendors keep making as they chase enterprise.
3. Plan for continuity in your pricing narrative. For “trustworthy AI” tooling, the shutdown showed that switching-cost and exit risk are part of the real price buyers pay — relevant for anyone pricing mission-critical infra in the Infrastructure, Compute & MLOps category.

Sources

• Humanloop homepage — “Humanloop is joining Anthropic” notice (accessed 2026-06-09)
• Humanloop pricing page — final public state, free trial + Enterprise (accessed 2026-06-09)
• TechCrunch — Anthropic nabs Humanloop team (Aug 13, 2025) (accessed 2026-06-09)
• Nordic9 — Humanloop to raise $2.6M (Index Ventures, YC, LocalGlobe, AlbionVC) (accessed 2026-06-09)
• SaaSworthy — Humanloop pricing (third-party listing, historical $ tiers) (accessed 2026-06-09)

Bottom line

Humanloop was a credible enterprise LLM-evals, prompt-management and observability platform that priced the workflow layer on a log/datapoint meter rather than reselling tokens — a clean value-metric choice. But in August 2025 Anthropic acqui-hired its founders and team without buying the IP, and Humanloop sunset the platform through late 2025. The pricing on this page is a post-mortem: historical and indicative, no longer purchasable. Browse the pricing blueprint for live, fully-researched company profiles, or compare against other LLMOps and evals tools there.

Want to compare Humanloop against other Infrastructure, Compute & MLOps companies? Browse the pricing blueprint.