How we work

Methodology

How LLMDeals researches, verifies, and ranks AI deals.

Our process

1
Discovery
We monitor provider websites, pricing pages, and announcements for new deals, changes, and expirations.
2
Verification
Every fact is traced to an official source. We capture evidence with URLs and timestamps.
3
Normalization
We convert heterogeneous pricing models into comparable deal routes (offer × model combinations).
4
Assessment
We evaluate each deal route on economics, model quality, usability, and evidence confidence.
5
Editorial
We write plain-language explanations of why each deal matters and who it's best for.

Consumer ratings

Every deal gets a rating based on its overall value proposition:

ExceptionalOutstanding value, highly recommended
ExcellentStrong value, worth using
GoodSolid option, meets needs
NicheGood for specific use cases

Deal types

We categorize deals into simple consumer labels:

FreeGenuinely useful recurring free access
SubscriptionPay monthly for discounted usage
PromoTemporary discount or credit
UnlockPayment unlocks benefits

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