#2 in llm gateways · mit-licensed, and billed per log rather than per token
87/100
verdictthe only gateway here with a complete feature set under a genuinely permissive licence, priced in a unit that decouples your gateway bill from your token spend.
Portkey vs Helicone
Helicone
Portkey
price
$79/mo (pro)
$49/mo (production)
free tier
yes
yes
markup on tokens
0% — byok supported
none disclosed
platform fee
$79/mo (pro)
$49/mo (production)
self-host
yes — apache 2.0
yes — mit
models carried
100+ providers
1,600+ claimed
switch forteams who want the full governance stack and the option to run it themselves
pros
+mit licence with no enterprise carve-out — genuinely self-hostable
+gateway bill is independent of token volume
+fallbacks, retries and timeouts on every tier including free
+virtual keys with budgets for per-team spend control
+guardrails documented rather than sold as an enterprise conversation
cons
−per-log billing is hard to forecast for high-request workloads
−free tier keeps logs for three days, which is too short to debug with
−publishes a broken '0+ llms' stat next to its 1,600+ claim
#6 in llm gateways · the widest provider coverage, with the governance behind a paywall
78/100
verdictthe broadest coverage of any self-hostable gateway, but the features most teams deploy it for sit under a proprietary licence with no published price.
LiteLLM vs Helicone
Helicone
LiteLLM
price
$79/mo (pro)
free to self-host — enterprise is quote-only
free tier
yes
yes
markup on tokens
0% — byok supported
none — self-hosted
platform fee
$79/mo (pro)
quote only
self-host
yes — apache 2.0
yes — mit core, proprietary enterprise/
models carried
100+ providers
100+ providers
switch forteams who want one interface to every provider and can live without the enterprise tier
pros
+100+ providers behind one openai-format interface
+virtual keys and per-key spend caps on the free mit core
+no per-token cost when self-hosted — provider bills direct
+load balancing, retries and fallback routing built in
+licence is explicit about where the commercial line falls
cons
−sso, rbac, audit logs and multi-tenancy are all enterprise-licensed
−enterprise pricing is quote-only, sold per seat
−the open-core split means an unlicensed production rollout is easy to stumble into
#7 in llm gateways · 5% on model spend, stated plainly and applied consistently
75/100
verdictthe most transparent pricing in the category and the only one that charges for inference, which makes it costlier than the pass-through gateways on identical traffic.
Requesty vs Helicone
Helicone
Requesty
price
$79/mo (pro)
$0 free tier — pay-as-you-go is 5% on model cost
free tier
yes
yes
markup on tokens
0% — byok supported
5% on model cost
platform fee
$79/mo (pro)
none
self-host
yes — apache 2.0
no
models carried
100+ providers
600+ / 300+ — conflicting
switch forteams who would rather pay one legible percentage than model a platform fee
pros
+one legible number — 5% on model spend, no seats or minimum
+eu data residency available on every plan
+fallbacks and spend limits included at the first paid tier
+routing and caching on the free tier
+openai-compatible endpoint with a documented base url
cons
−the only gateway here that marks up inference at all
−guardrails, pii detection and sso are enterprise-only at custom pricing
−publishes two different model counts on its own properties
#8 in llm gateways · apache 2.0 on paper, open-core in the build
72/100
verdictthe fastest thing here and the most permissively licensed on paper, but the features that make a gateway safe in production are not in the apache build.
Bifrost vs Helicone
Helicone
Bifrost
price
$79/mo (pro)
free (self-host) — enterprise is custom
free tier
yes
yes
markup on tokens
0% — byok supported
none — self-hosted
platform fee
$79/mo (pro)
none (oss)
self-host
yes — apache 2.0
yes — apache 2.0
models carried
100+ providers
23+ providers
switch forlatency-sensitive proxying where the gateway must not be the bottleneck
pros
+unmodified apache 2.0 licence with no carve-out in the repository
+roughly 20µs added latency, the lowest published here
+drop-in replacement for the openai, anthropic and google sdks
+hierarchical budgets across virtual keys, teams and customers
+runs from a single go binary with zero-config startup
cons
−guardrails, clustering and the mcp gateway are enterprise-only despite the licence
−provider and model counts contradict between repo and product page
−enterprise pricing is quote-only with no published figure
every tool on this page went through the same test as Helicone — same tasks, same order, scored the same way. the comparison tables are the figures from that testing, not vendor spec sheets.