How is inference billed on Helios Labs?

Inference is billed per million tokens processed, with separate line items for input and output tokens. Rates vary by model class — for example, 70B-parameter models on H100 clusters are billed at the published Llama-3-70B rate shown on the Team and Enterprise tier cards. Cached prompts and routing misses are itemized transparently on your monthly invoice.

What counts as an 'evaluation run' for billing?

An evaluation run is one execution of an eval suite against a model checkpoint or deployed endpoint, regardless of dataset size. Each invocation counts as one run, and runs are priced per 1,000 executions per month. Failed runs under 30 seconds are not billed, so retry logic doesn't inflate your invoice.

Is there a free trial or self-serve Developer tier?

Yes. The Developer tier is free up to 500,000 inference tokens and 100 evaluation runs per month, with no credit card required. It includes the full SDK, the observability dashboard, and community Discord support. Pro teams typically graduate to the Team tier once they exceed those limits.

Do you offer annual commits or volume discounts?

Annual commits receive a 15–22% discount depending on committed spend tier, applied as contractually guaranteed minimums with overages billed at the published rate. Teams forecasting $50K+/month should request a custom quote through the demo flow — our solutions engineering team will model your workload before quoting.

Can we deploy Helios into our own VPC or on-prem H100 cluster?

Yes. Enterprise customers can run Helios as a dedicated tenant in our SOC 2 Type II environment, in a Helios-managed VPC inside your AWS or GCP account, or as a fully self-hosted install on your own H100 cluster. Self-hosted deployments use the same pricing model with a negotiated infrastructure pass-through.

How do we get SOC 2 and HIPAA documentation for procurement?

Enterprise prospects can request the SOC 2 Type II report and HIPAA attestation under NDA within 24 hours of booking a demo. Our Trust Center also publishes subprocessor lists, data flow diagrams, and penetration test summaries. Procurement and security teams are encouraged to join the demo so we can answer architecture questions live.

Pricing / Usage-based

Pricing built like the runtime: per-token, per-eval, no surprises.

Three tiers, transparent unit economics, and a live estimator so you can model the line items before you talk to anyone. When your usage moves into enterprise territory — custom MSAs, SOC 2 paperwork, BYOC — the fastest path is a 30-minute call with solutions engineering.

Book a 30-minute demo Run the estimator →

SOC 2 Type IIsince Mar 2024
2.1MSDK downloads
99.97%trailing uptime
a16z-led$19M Series A

helios-estimate --tokens 240M

# monthly cost projection
workload      "chat + eval (llama-3-70b)"
tokens        240_000_000      // 240M / month
inference     $1,182.40         // $0.0000049 / 1k tok
evals         $  186.00         // 142 runs * $1.31
observability $  248.00         // 4 regions, 30d ingest
seats         $    0.00         // 8 eng seats, included
---
subtotal      $1,616.40         // est. bill
tier          "Team"            // unlocks at 250M

// > 250M tokens / month? Use the demo button —
//   custom rates + annual commit available.

estimator live · v0.4.2 output refreshed in 0.31s

Tier matrix

Three tiers. One runtime. Pick the row your GPUs would.

Developer is free up to a working monthly quota. Team is the default for production teams shipping LLM features in earnest. Enterprise is a contracted rate card with audit, procurement, and deployment knobs you actually need.

✓Up to 1M tokens / month free — inference + eval combined quota
✓2 seats included, additional seats $29 / seat
✓Community support, GitHub-issues SLA, public status page
✓Single region (us-west-2), 7-day metric retention
✓Pay-as-you-go overage: $0.0000049 / 1K inference tokens
×SOC 2 audit pack
×BYOC / hybrid deployment

Start free →

Most teams land here

✓250M tokens / month included, $0.0000049 / 1K overage
✓10 seats included, $49 / seat after; SSO & SCIM
✓5,000 eval runs / month, $1.31 per additional run
✓Multi-region (us, eu, ap), 30-day retention
✓Shared Slack channel, 8-business-hour response SLA
✓SOC 2 Type II report under NDA in 24h
✓HIPAA-ready configuration available

Book a demo →

✓Volume-priced inference, locked unit rates for the contract term
✓Unlimited seats & regions, custom retention windows
✓BYOC / hybrid deployment on your VPC, our control plane
✓Custom MSA, DPA, BAA, security review support
✓Named solutions engineer, 99.95% uptime SLA, 1h P1
✓Annual commit discount up to 18% on blended usage
✓HIPAA, PCI, FedRAMP-ready deployment paths

Talk to solutions engineering →

Inference unit rates $0.0000049 per 1K tokens · $1.31 per eval run · $0.18 per region-day Same unit economics on every tier — only the included quota and support ceiling change.

Cost estimator

Estimate your monthly bill in four inputs — no email required.

Use the worked example as a sanity check, then plug in your own numbers. The estimator uses the same rate card our billing engine uses — if the math here surprises you on a demo call, that’s our problem, not yours.

Monthly tokens

Sum of prompt + completion tokens across all inference calls in a 30-day window. The estimator auto-bins you into the right tier once you cross 250M.

Example 240,000,000 tokens — 12M prompts avg × 20 calls × 1k req/day

Model class

Helios prices inference by token, but model class affects the latency tier and whether eval pricing is per-run or per-step.

Example llama-3-70b on H100 — $0.0000049 / 1K tokens, p99 < 180ms

Eval runs / month

Each eval run bundles scoring + judge-model calls + trace capture. 5,000 runs are bundled on Team; above that it’s $1.31 per run.

Example 142 eval runs × $1.31 = $186.02 (all on Team quota)

Regions & retention

Active regions drive observability ingest at $0.18 / region-day. Retention over 30 days is $0.0042 / GB-day on cold storage.

Example 4 regions × 30 days = $21.60/day ingest baseline

Worked example · 240M tokens / month, llama-3-70b

How the line items stack

Inference 240M tokens × $0.0000049 / 1K $1,182.40

Eval runs 142 runs, all inside Team quota $186.02

Observability ingest 4 regions × 30 days × $0.18 $21.60

Seats 8 engineers, all inside Team quota $0.00

Estimated monthly bill Team tier · pre-annual-commit $1,616.40

Annual commit on Team unlocks a 12% blended discount — bring your forecast to the demo and we’ll model it live.

If your forecast exceeds 250M tokens / month

You’re an Enterprise conversation. Let’s get on a call.

Book a 30-minute demo

Consolidation math

Four vendors → one invoice. Show the line items.

The 2024 MLOps Community Migration Report verified that Helios replaces W&B + LangSmith + BentoML + Helicone for teams running LLM workloads. Here’s what a typical mid-market team was actually paying, vs. the unified Helios bill.

Weights & Biases — team plan$1,650
LangSmith — Plus seats$1,470
BentoML — inference cluster$2,840
Helicone — observability$640
Glue engineering (FTE fraction)$1,800
Vendor management overhead$300

Stack total / month $8,700

Plus 4 procurement cycles, 4 security reviews, 4 on-call rotations.

Inference — 240M tokens @ $0.0000049 / 1K$1,182.40
Evaluations — 142 runs$186.02
Observability — 4 regions$21.60
Seats — 8 engineers, included$0.00
Glue engineering$0
Vendor management overhead$0

Helios total / month $1,390

One invoice. One security review. One on-call rotation.

4 → 1

vendors consolidated

84%

lower monthly run-rate

4.7×

faster ship velocity (verified)

24h

SOC 2 report turnaround

Procurement FAQ

The questions your finance team will ask before the call.

Written so your controller and your CISO can pre-validate without a synchronous meeting.

How do overages work, and what happens when we burst past our tier quota?

Overage is metered at the same per-token and per-eval-run rate you see on this page — no surge multipliers, no “contact sales” thresholds. We send a soft alert at 80% of monthly quota and a hard alert at 100%. You can set a hard spend cap per workspace and pause inference automatically when it trips. Team tier includes 250M tokens / 5,000 eval runs; usage above that is billed monthly in arrears.

What discounts are available for annual commits, and how is the commit sized?

Annual commits are sized off your highest-traffic 30-day window from the trailing 90 days, with a 1.2× multiplier for headroom. Discounts scale with commit size — 8% on Team-tier commits, up to 18% on Enterprise blended commits above 5B tokens / year. Unused commit can be rolled forward one billing cycle; overage beyond commit is billed at standard rates.

Can we deploy Helios inside our own VPC or on-prem, and what does BYOC actually mean?

BYOC (Enterprise tier) runs our control plane in our SOC 2 boundary and the inference + observability data plane inside your AWS, GCP, or Azure VPC. Your data never leaves your account; Helios operators access it via short-lived, just-in-time credentials. Hybrid is also supported — control plane in our tenant, data plane on-prem on your H100s. Typical provisioning is 2–3 weeks for a standard VPC layout.

Is HIPAA available, and what about custom MSAs, DPAs, and BAA paperwork?

HIPAA attestation was completed in November 2024 and is available on Team and Enterprise tiers via a signed BAA — typically a 3-business-day turnaround once redlines are exchanged. SOC 2 Type II reports (since March 2024) are available under NDA within 24 hours via the trust portal. Custom MSAs, DPAs, and security addenda are handled by solutions engineering and our standard redline-to-sign turnaround is 5 business days.

Still need to validate something specific for your security review?

Book a 30-minute demo

Solutions engineering

Need scale, SOC 2 paperwork, or a custom MSA? Talk to solutions engineering.

If your forecast pushes past 250M tokens / month, if you need a BAA, or if procurement needs a custom MSA — skip the pricing back-and-forth. 30 minutes, a screen share, and a real engineer who has shipped this before.

Book a 30-minute demo [email protected]

Solutions engineer responds within 1 business day
SOC 2 report under NDA within 24h
Custom MSA redline turnaround in 5 business days