Leanroute vs OpenRouter

Flat BYOK subscription, APAC depth, Singapore residency.

OpenRouter is the category leader: 400+ models, mature ecosystem, live since 2023. We launched in 2026 with a specific bet — a $15/$25 flat BYOK subscription that eliminates per-request fees entirely (OpenRouter can't match this without ripping up their revenue model), deeper APAC + India provider catalog, and Singapore residency. If you're building for Asia, your monthly spend is above ~$500, or you need a Singapore-incorporated vendor, the math tilts our way. If you need a model that's only on OpenRouter's long tail, you value seniority + community size, or you're well below $500/mo of usage, theirs is the safer pick.

Snapshot date: July 2026. Pricing and feature claims re-validated quarterly — check openrouter.ai before depending on a specific number.

TL;DR

Pick Leanroute if (a) you route to APAC or Indian providers, (b) Singapore residency is a hard requirement, (c) your monthly upstream spend is above ~$500 (the $15 Starter or $25 Pro subscription pays back many times over vs. a per-request fee), or (d) you already have provider accounts and want observability + failover without another metered fee. Pick OpenRouter if (a) you need a model that lives only in their long tail, (b) you value vendor seniority, (c) you're US-resident and don't want to think about residency at all, or (d) your monthly spend is small enough that any flat subscription costs more than metered usage.

Where we differ

Dimension	Leanroute	OpenRouter
Fee on managed-key requests	5% at top-up (once), $0 per request	5% per request (compounded)
Fee on BYOK requests	$0 per request ($15/$25/mo subscription covers it)	5% per request (BYOK is paid tier)
Cache hit cost	$0	Full fee
Subscription / minimum	Optional BYOK subscription ($15 Starter / $25 Pro) or prepaid credits ($100 min card)	None for routing; paid tiers for higher limits
Total models in catalog	~38 (curated)	400+ (long tail)
APAC providers (Qwen, GLM, Doubao, Kimi)	First-class, vision included	Partial (Qwen yes; others limited)
India providers (Sarvam, Krutrim)	Yes, both	No
Primary data residency	Singapore	US
Streaming + tool calls (every provider)	Yes (#25)	Yes for top providers; gaps in long tail
Per-org cache isolation	Yes, namespaced	Yes
No-persistence mode	Per-org toggle + verification header	Account-level setting
Multi-account failover	Yes (5xx → same-tier alt, #23)	Yes (similar mechanic)
Spend caps / rate limits	Org + per-key, hard 402/429	Account level
Track record	Launching mid-2026	Live since 2023

Rust-tinted cell: Leanroute leads. Muted cell: OpenRouter leads. No tint: tie or context-dependent.

The cost math

Upstream cost / month	Leanroute (best pick)	OpenRouter (5%)	You save
$100	$105 (credits)	$105	~$0 (parity)
$500	$515 (BYOK Starter)	$525	$120 / yr
$1,000	$1,015 (BYOK Starter)	$1,050	$420 / yr
$10,000	$10,025 (BYOK Pro)	$10,500	$5,700 / yr
$100,000	$100,025 (BYOK Pro)	$105,000	$59,700 / yr

Under $500 / month of upstream spend, our credits plan (5% at top-up) sits at parity with OpenRouter's 5% per-request fee — you're paying the same in aggregate, we just take the fee once at deposit rather than compounded on every call. The math tilts hard as soon as you cross the subscription break-even point: at $1,000/mo the $15 Starter subscription saves $35/month vs OpenRouter's $50, at $10,000/mo the $25 Pro subscription saves $475/month ($25 flat vs $500 metered), and at $100,000/mo Pro saves $4,975/month ($25 flat vs $5,000).

Break-even is where the $15 Starter subscription equals OpenRouter's 5% cut: $15 / 0.05 = $300 of monthly upstream. Above $300/mo, Starter wins. Above ~$500/mo (when Starter's 1M request cap starts biting on light-token traffic), Pro takes over and wins by an expanding margin forever after.

Cache hits widen the gap further. We bill them at $0; if your workload is 40% cache-served (typical for RAG + system prompts), 40% of your bill drops to $0 with us versus (upstream + 5%) with them — on a $1,000 workload that's an additional $20/mo saved even on the pay-as-you-go credit plan.

Run the numbers yourself at /calculator.

Provider catalog

OpenRouter has more models in absolute terms (400+ vs our 38). That number is real and meaningful when you need a specific fine-tune that lives in their long tail. For the top 20 models that 80% of production traffic actually uses, we cover the same ground — OpenAI / Anthropic / Google / DeepSeek — with one curated entry each rather than a dozen variants.

Where the catalogs diverge meaningfully is APAC. We treat Qwen (Qwen-Max, Qwen-VL), GLM (4-Plus, 4V-Plus), Doubao (1.5 Pro, Vision Pro), Kimi (Moonshot V1) as first-class with vision routing, not as long-tail listings. We also list Sarvam (sarvam-30b, sarvam-105b) and Krutrim (Spectre V2, Krutrim 2) which OpenRouter doesn't carry at all. If you're building Hindi / Tamil / Mandarin features, that's the difference between "works" and "doesn't."

Full catalog at /docs/models.

Data residency

We're a Singapore Pte. Ltd. with primary hosting in Singapore — compute, database, and cache all in-region. OpenRouter routes through US edges. Both options are valid; the question is whether your compliance posture, your customers' expectations, or your latency from APAC endpoints makes one a hard requirement.

Full posture against PDPA / PDPO / GDPR / PIPL / APPI / DPDP at /compliance.

Where OpenRouter still wins

Model breadth. 400+ vs 38. If your workload depends on a specific fine-tune (Nous Hermes, MythoMax, a community Mixtral variant), they carry it and we don't.
Track record. Live since 2023, much bigger user community, more third-party integrations have OpenRouter-specific quickstarts.
US presence. If your buyers want a US-incorporated vendor with US hosting, that's us missing the brief.
Prompt routing UI. They have a more polished chat playground for ad-hoc model testing; ours is API-first.