Our process
Consumer ratings
Every deal gets a rating based on its overall value proposition:
Deal types
We categorize deals into simple consumer labels:
Ranking formula
Internal scores combine multiple factors:
rank_score = 0.35 economic_advantage + 0.30 model_capability + 0.20 usability + 0.10 reliability + 0.05 evidence_confidence
Penalties apply for short expiry, severe rate limits, regional restrictions, difficult signup, dynamic/unknown quotas, and weak evidence.
Effective value multiple (the honest metric)
A subscription's headline "value" is meaningless until you account for how much you can actually use. We compute it per model:
agent request profile: 830 input + 71,500 cached + 295 output tokens cost_per_request = (in*830 + cached*71,500 + out*295) / 1M [per-model rates] effective_multiple = monthly_request_cap x cost_per_request / subscription_fee
Above 1x beats pay-as-you-go. Below 1x, the subscription destroys value for that route — we flag it in red rather than hide it.
Mega-deal detection
Adapted from our dell research stack. A route is flagged 🔥 Mega only when stated facts cross thresholds:
- Request quota ≥ 100K/month
- Effective multiple ≥ 2x (≥3x scores higher)
- Free tier with ≥1,000 requests/day
- Any route below 1x is disqualified regardless of other factors
Every badge shows its reasons — no opaque single scores.
Task-profile scores
The homepage table rescores every route under four weightings:
overall: quality .35 / cost .35 / reliability .20 / throughput .10 coding: quality .40 / cost .30 / reliability .20 / throughput .10 agents: quality .30 / cost .25 / reliability .25 / throughput .20 research: quality .35 / cost .25 / reliability .20 / throughput .20
What we don't do
- We don't rank by affiliate commission
- We don't certify providers
- We don't claim a lower price automatically means a better route
- We don't invent quotas or pricing we can't verify