OpenRouter Pricing 2026: Complete Model Cost Guide & Hidden Markup Breakdown
TL;DR
The platform maintains "no markup on inference" with per-token rates matching published provider cost. However, teams encounter a 5.5% credit-card platform fee (with $0.80 minimum floor) and a 5% BYOK charge on requests exceeding 1M monthly. Total overhead: plan for 5-7% above stated model costs.
OpenRouter's Real Fee Structure
Three distinct billing layers apply to all spending:
| Layer | Fee | Trigger | Source |
|---|---|---|---|
| Platform fee (credit card) | 5.5%, $0.80 minimum | Every top-up with card | openrouter.ai/pricing |
| Platform fee (crypto) | 5% | Every crypto top-up | Simplifying platform fee announcement |
| Inference (per token) | Passthrough at provider rate | Every API call | openrouter.ai/pricing |
| BYOK overage | 5% of equivalent OpenRouter price | Requests after 1M/month with own key | 1 million free BYOK announcement |
The inference passthrough receives marketing emphasis—and it's legitimate. Platform fees generate actual revenue but receive less prominent disclosure.
The $0.80 Minimum Impact
The credit-card fee carries a floor that disadvantages small-scale usage. Depositing $5 incurs $0.80 in fees (16% effective rate), while a $10 deposit means $0.80 remains (8% effective rate). The nominal 5.5% rate only applies around $15+ deposits. Consolidating to $50-$100 loads normalizes the percentage.
Claude Markup Claims Are Outdated
Earlier analyses cited a 100% Claude markup ($6/$30 versus $3/$15 direct). As of May 2026, Claude Opus 4.7 rates match Anthropic's published pricing at "$5 per million input tokens, $25 per million output." Similar parity holds for Sonnet 4.6 ($3/$15) and Haiku 4.5 ($1/$5). The discrepancy reflects historical pricing, not current practice.
OpenRouter routes across multiple providers per model. Rate changes by any provider update the platform's listed rate accordingly. This ensures "no markup," though not necessarily "lowest available" pricing.
BYOK Free Tier and Overage Structure
October 2025 introduced 1 million free BYOK requests monthly. Exceeding this ceiling triggers a 5% charge calculated from equivalent OpenRouter costs, despite direct provider billing for tokens themselves.
Scaling example:
- 1,000 active developers
- 20 agent runs per developer daily
- 4 LLM round-trips per run
- = 80,000 requests/day
- = 2.4M requests/month
This scenario places 1.4M requests above the free tier. The fee scales proportionally with usage—no flat overage cap exists.
Mitigation strategies: Stay under 1M through aggressive caching and batching, or accept the fee as unified routing's cost.
Undisclosed Additional Costs
Stripe FX Conversion (Non-USD Cards)
Non-USD card top-ups incur 1-3% conversion markup via Stripe before the 5.5% platform fee applies. Total: approximately 7-8% for EUR or JPY deposits. The marketing dashboard won't reflect this—verify actual received credits against bank statements.
Failed Routing Retries
Provider failover routes traffic when errors occur. Certain providers charge partial tokens before failing, meaning both attempts get billed. The aggregate charge appears as a single line item. This typically represents less than 0.5% monthly overhead and remains documented in the FAQ, though billing aggregation lacks clarity.
Free Model Rerouting
OpenRouter lists "free" open-source variants (Llama 4 Scout, DeepSeek V4 Flash, Qwen3 Coder, Gemma 3) with sponsored provider backing and strict rate limits (typically 20 req/min, 200 req/day). When limits exhaust, traffic may queue, drop, or silently reroute to paid providers at full rates if fallback configuration applies. Production workloads on free routes should monitor actual provider responses—cost differences between free and fallback paid can reach 100x.
Real-World Cost Scenarios
Scenario 1: Solo developer, $50/month Claude usage
- Token cost: $50.00
- Platform fee (5.5%): $2.75
- Total: $52.75
- Effective overhead: 5.5%
This represents OpenRouter's core value proposition—convenience at a modest premium.
Scenario 2: Indie developer, five models at $5 per top-up
- Five $5 deposits with $0.80 minimum each: $4.00 fees
- Token usage: $25.00
- Total: $29.00
- Effective overhead: 16%
The lesson: consolidate funding to $25+ single deposits.
Scenario 3: Startup (50 developers), BYOK with Anthropic, under 1M requests
- Total requests/month: ~600,000
- BYOK fee: $0
- Anthropic token cost: $4,200/month
- OpenRouter cost: $0/month
- Total: $4,200/month
Here OpenRouter provides routing and observability at zero marginal cost.
Scenario 4: Same startup at 3x growth
- Total requests/month: 1.8M
- Average equivalent OpenRouter price per request: ~$0.008
- BYOK fee (800K over limit × $0.008 × 5%): $320/month
- Anthropic token cost: $12,600/month
- Total: $12,920/month
When OpenRouter Wins and Loses
Advantages:
- Multiple model access through one API key
- Sufficient top-up volume to amortize the $0.80 minimum
- BYOK usage under 1M/month
- Mid-sized teams requiring Claude + GPT + Gemini parallelization benefit from unified relationship management at a reasonable 5.5% premium.
Disadvantages:
- Single standardized provider with negotiated rates pays OpenRouter overhead on top
- Small frequent top-ups (victims of the $0.80 floor)
- BYOK scaling past 1M/month without awareness
- Absolute cost-per-token priority favoring discount aggregators or direct provider access
OpenRouter optimizes for breadth and ease. Cost-minimization strategies typically require alternatives.
Comparison Alternatives
ofox.ai — Single-key, OpenAI-compatible aggregator offering automatic volume-tier discounts (Bronze/Silver/Gold/Platinum providing 3–7% off at $1K/$5K/$10K/$20K monthly thresholds). No credit purchase fee; charges reflect actual usage. Covers Claude Opus 4.7/4.6/Sonnet 4.6, GPT-5.5/5.4 families, Gemini 3.5/3.1, DeepSeek V4 Pro/Flash, Grok 4.20, Qwen 3.7 Max, Kimi K2.6.
Direct provider accounts — Per-token minimum cost with zero overhead. Trade-off: engineering effort maintaining multiple billing relationships.
Cloud-vendor inference (AWS Bedrock, GCP Vertex) — Includes Claude/Llama/etc. with cloud-provider billing mechanics. Typically 10-20% more expensive per token but integrates existing cloud spend.
Verification Methodology
Pricing evolves—validate independently:
- Extract current rates from openrouter.ai/pricing and compare against provider published prices.
- Execute a $10 test top-up; verify actual received balance. The difference represents real platform fees.
- Make 10 actual calls and inspect response headers showing which provider executed. Calculate effective cost from billing dashboard versus quoted rate.
- BYOK users should query
/api/v1/usageendpoint to track request count toward 1M/month ceiling. Set alerts at 800K threshold.
Bottom Line
OpenRouter's pricing operates transparently yet invites misreading. The inference passthrough (genuinely zero markup) dominates marketing messaging while platform fees (equally genuine, non-zero) receive secondary emphasis. Budget 5-7% overhead on credit-card spending, monitor BYOK request counts approaching the 1M threshold, avoid sub-$20 deposits when possible, and recognize "no markup" and "no cost" as distinct claims. Costs manifest in different line items than direct provider access, not necessarily lower totals.
Originally published on ofox.ai/blog.
Top comments (0)