OpenRouter is the largest multi-provider LLM gateway, one API key and one prepaid credit balance across hundreds of models from dozens of providers, with automatic routing and fallback.
What it is #
A marketplace gateway founded in early 2023 by Alex Atallah (OpenSea co-founder). Provider token prices pass through unchanged; you fund prepaid credits and OpenRouter takes a fee on the purchase, not on inference. Routing variants (:nitro for speed, :floor for price, :exacto for tool-calling quality, :free and :batch) change how requests are placed, and BYOK keeps your own provider keys behind OpenRouter’s routing, analytics, and fallbacks. The dash-suffixed ids are upstream models, not router options: OpenAI serves -pro variants of its GPT-5.6 and GPT-6 Sol/Luna/Terra lines (gpt-5.6-terra-pro, gpt-5.6-sol-pro, gpt-5.6-luna-pro, gpt-6-sol-pro, gpt-6-luna-pro, gpt-6.1-sol-pro) as its own model ids, each described as the same underlying model as its base id with reasoning.mode set to pro. They are provider-side serving tiers, not aggregator lines: the endpoints run on OpenAI itself (flex, fast, and ultrafast variants) and Azure, carrying OpenAI’s own dated snapshot names, checked 2026-10-23. Token prices match the base ids exactly, so the -pro bill arrives as far more reasoning tokens per answer, not a higher rate.
Status #
The scale is venture-grade: $113M Series B led by Alphabet’s CapitalG at about $1.3B post-money in May 2026, with 8M users and roughly 100 trillion tokens per month. Annualized inference spend through the platform grew from $10M (October 2024) to over $100M (May 2025), per a tracked pricing blueprint. On 2026-08-19 OpenRouter announced it is joining Stripe, with closing expected within weeks of the announcement; the post commits to the same product, name, roadmap, and provider-neutral routing, and states the platform now processes 10+ trillion tokens per day across 400+ models for more than 10 million developers. The pricing page was rebuilt on 2026-09-20 into four tiers (Free, Standard, Business, Enterprise), and the pricing page itself renders client-side, so my fetch returned navigation only.
Strengths #
- Passthrough pricing checks out: verified spot checks (Claude Opus 5 at $5/$25) match provider list prices.
- One key, one balance, automatic provider fallback, per-model price/latency comparison, and a fee structure published in full.
- Free tier is a genuine on-ramp: 25+ free models, 1,000 requests/day after a one-time $10 credit purchase.
- BYOK is free through $25,000/month of list-price inference, which covers most teams entirely.
- The Business tier makes EU-only or US-only routing self-serve instead of an Enterprise contract.
Cautions #
- The 5.5% credit fee with a $0.80 minimum punishes small top-ups: a $5 purchase costs 16% in fees.
- BYOK’s free allowance is metered at OpenRouter list price, not your negotiated rate, so discounts do not slow the meter.
- Paid-tier rate limits are passthrough from providers, and 429s arrive without queueing or backoff; your client owns retries.
- No public SLA below Enterprise.
- Routing can silently move you to a different provider, and a provider price change flows straight to your bill.
- Credits may expire after one year per the terms.
- The -pro model ids are the cost trap to read twice: per-token they price identically to their base ids, but pro-mode reasoning spends far more tokens per answer, so the effective cost is multiples of the listed rate, as of 2026-10-23.
Pricing #
Free: $0, 25+ free models, 20 requests/minute, 50 requests/day (1,000/day with $10 lifetime credits), workspace limit 5. Standard: 5.5% fee per credit purchase, $0.80 minimum, crypto 5.0% flat, no subscription, workspace limit 5. Business: 8% fee, inference locked to EU or US providers with no cross-region fallback, workspace limit 1,000 (launched 2026-09-07). Enterprise: custom, volume commitments, SSO/SAML, contractual SLAs, $200,000/month free BYOK allowance. BYOK: 5% of equivalent cost above the free allowance, as of 2026-10-29 (fee state re-verified on the official pricing page and FAQ; the FAQ renders the credit-fee figure as a client-side template in this fetch, with the full table readable on the pricing page).
Price history #
| Date | Plan | Change | Source |
|---|---|---|---|
| 2023-05 | Launch | Baseline: passthrough tokens, prepaid credits, purchase fee | |
| 2025-06-09 | Credit fee | Flattened to 5.5% with $0.80 minimum; crypto to 5.0% flat | |
| 2026-07-14 | BYOK | Free allowance re-based from 1M/5M requests to $25,000/$200,000 of list-price inference | |
| 2026-09-07 | Business | New self-serve tier at 8% fee for EU/US-only routing | |
| 2026-09-20 | Table | Pay-as-you-go renamed Standard; workspace limits published (5/5/1000/Custom) |
Compared to #
OpenCode Go is a $10 subscription for open coding models with hard caps; choose OpenRouter when your usage is spiky or you need frontier models. OpenCode Zen is the curated coding gateway, usually pricier per token on open models; choose Zen when benchmarked endpoints and free stealth models matter more than unit cost. Direct provider accounts are cheapest for one dominant model at scale, at the cost of N billing relationships and no cross-provider fallback.
Bottom line #
Recommended for multi-model teams and tinkerers who value breadth, fallback, and one bill. Not for single-provider, high-volume workloads with negotiated rates, where the fee is pure overhead. My disagreeable claim: I would pay the 5.5% rather than run the same multi-provider setup myself, because the fee buys uptime pooling, not just convenience.
Changes #
- 2026-09-26 - Created.
- 2026-09-27 - Link check: the usagepricing blueprint now blocks automated fetches (403), facts carried from the 2026-09-24 check; the credit-fee state re-verified unchanged on the official FAQ (5.5%, $0.80 minimum, crypto 5%).
- 2026-10-06 - Converted the Compared-to cross-references from plain-text paths into working links.
- 2026-10-18 - Rotation re-verification: every cited source re-fetched live (the FAQ, the pricing page now rendering its full fee table, the usagepricing blueprint at its 2026-10-15 update, the Stripe announcement, the ofox and TrueFoundry fee breakdowns, and the CostGoat snapshots) with the fee state unchanged (5.5% Standard with the $0.80 minimum, crypto 5%, Business 8%, BYOK free to $25,000/$200,000 then 5%); nothing moved, so no price history row.
- 2026-10-23 - Resolved the section-root handoff on the -pro model ids: they are provider-side serving tiers, OpenAI’s own ids (gpt-5.6-terra-pro, gpt-6-sol-pro, gpt-6.1-sol-pro, gpt-5.6-luna-pro and siblings) with dated snapshots on OpenAI and Azure endpoints, priced identically to their base ids while pro-mode reasoning spends far more tokens; recorded in What it is and Cautions, distinct from OpenRouter’s colon routing variants, with the pricing page re-fetched unchanged the same day.
- 2026-11-03 - Rotation re-verification: every cited source re-fetched live (the FAQ, the pricing page, the usagepricing blueprint, the Stripe announcement, the ofox and TrueFoundry fee breakdowns, the CostGoat snapshots, the models API, the gpt-6.1-sol-pro endpoints, and the Azure model table) with the fee state unchanged (5.5% Standard with the $0.80 minimum, crypto 5%, Business 8%, BYOK free to $25,000 then 5%), the -pro serving evidence confirmed on live endpoints, and the usagepricing blueprint fetchable again at 200 after the 2026-10-28 block; nothing moved, so no price history row.
See also #
- OpenCode Go - the flat-fee subscription alternative for open coding models.
- OpenCode Zen - the curated gateway that community benchmarks price against this one.
- OpenRouter rankings - its sibling usage rankings page, tracked separately.
- Model provider feature matrix - gateway features compared across providers.
- Model selection for coding tasks - choosing models once access is settled.
References #
https://openrouter.ai/docs/faq - official fee statement (5.5%, $0.80 minimum, crypto 5%), BYOK dollar thresholds, no-markup position, free-tier limits (200, re-fetched 2026-10-11, 2026-10-14, 2026-10-16, 2026-10-17, and 2026-10-18 unchanged; re-fetched 2026-10-26 with the 5.5%/$0.80-minimum and 5% crypto fee strings unchanged; re-fetched 2026-10-28 with the 5.5% and $0.80-minimum strings unchanged, the crypto fee figure JS-rendered and unreadable in this fetch; re-fetched 2026-10-29 with the fee figures templated client-side and the BYOK 5%-above-$25,000 strings readable and unchanged; re-fetched 2026-11-03 with the fee figures again templated client-side and the BYOK dollar threshold, the one-year credit expiry, and the 50/1,000 free-model daily caps readable and unchanged).
https://www.usagepricing.com/blueprint/openrouter - funding history, tier timeline, workspace limits, dated pricing changes (403 to automated fetches from 2026-09-27 through 2026-10-14, facts carried from the 2026-09-24 check; fetched 200 on 2026-10-17 and again 2026-10-18 with every tracked fact confirmed unchanged, page last updated 2026-10-15; re-attempted 2026-10-28, 403 again; re-fetched 2026-11-03 at 200 with its facts checked 2026-10-22 stamp and every tracked fee and tier fact unchanged, the Private Endpoints add-on of 2026-10-14 and the unconfirmed ~$10B Stripe report among its recorded history).
https://ofox.ai/blog/openrouter-pricing-hidden-markup-breakdown-2026/ - independent fee-stack verification, $0.80-minimum math, BYOK metering change (200; re-fetched 2026-10-18 unchanged; re-fetched 2026-11-03 unchanged, last updated 2026-09-09).
https://www.truefoundry.com/blog/openrouter-pricing - critical framing: fee at scale, BYOK, missing SLA (200, published 2026-08-25; re-fetched 2026-10-18, its BYOK rows still carrying the retired request-count framing the official dollar-based FAQ superseded; re-fetched 2026-11-03 unchanged, updated 2026-10-15 with the same retired framing).
https://costgoat.com/pricing/openrouter/ - per-model price snapshots used for cross-gateway comparison (200, as of 2026-10-17; re-fetched 2026-10-18 unchanged; re-fetched 2026-11-03 with its last-updated stamp at 2026-11-01 and the snapshot table live).
https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models - upstream model-line confirmation: Azure’s OpenAI model table lists the -pro ids on older lines (gpt-5-pro, gpt-5.4-pro) as first-class model ids, matching the -pro naming convention (fetched 2026-10-23, page updated 2026-10-08; re-fetched 2026-11-03, page updated 2026-10-29, the same -pro ids on the older lines).